diff --git a/.github/PULL_REQUEST_TEMPLATE/tier5-queries.md b/.github/PULL_REQUEST_TEMPLATE/tier5-queries.md
new file mode 100644
index 000000000..c75043597
--- /dev/null
+++ b/.github/PULL_REQUEST_TEMPLATE/tier5-queries.md
@@ -0,0 +1,39 @@
+<!--
+Tier 5.5 Externally-Authored Query Submission template
+See eval/CONTRIBUTING.md for the full workflow.
+-->
+
+## Summary
+
+Submitting **N** Tier 5.5 queries for BrainBench.
+
+- Author handle: `@your-handle`
+- File location: `eval/external-authors/your-handle/queries.json`
+- Queries authored fresh (not copy-pasted from a model output)
+- Slugs verified against `eval/data/world-v1/` (via `bun run eval:world:view`)
+
+## Checklist
+
+- [ ] `bun run eval:query:validate eval/external-authors/your-handle/queries.json` passes
+- [ ] At least 20 queries
+- [ ] Each query has either `gold.relevant` (with real slugs) or `gold.expected_abstention: true`
+- [ ] Temporal queries have `as_of_date` set (`corpus-end` | `per-source` | ISO-8601)
+- [ ] Phrasing is varied (not all the same template)
+- [ ] `author` field matches my handle
+
+## Phrasing variety (optional self-audit)
+
+Tick the styles represented in your batch:
+
+- [ ] Full sentence questions
+- [ ] Fragment-style ("crypto founder Goldman Sachs background")
+- [ ] Comparison ("X vs Y")
+- [ ] Follow-up ("And who else...")
+- [ ] Imperative ("Pull up Alice Davis")
+- [ ] Trait-based ("the demanding engineering leader")
+- [ ] Abstention bait (answer is "not in corpus")
+
+## Notes to reviewer
+
+Anything worth flagging — ambiguous cases, corpus gaps you found, specific
+phrasings you were uncertain about.
diff --git a/.gitignore b/.gitignore
index 229382f0c..fc91e5fa3 100644
--- a/.gitignore
+++ b/.gitignore
@@ -13,3 +13,7 @@ supabase/.temp/
 .claude/skills/
 .idea
 eval/reports/
+eval/data/world-v1/world.html
+
+# BrainBench amara-life-v1 Opus cache (regenerate via eval:generate-amara-life)
+eval/data/amara-life-v1/_cache/
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 96359198a..559c33f45 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,67 @@
 
 All notable changes to GBrain will be documented in this file.
 
+## [0.20.0] - 2026-04-23
+
+## **BrainBench moves out. gbrain gets its install surface back.**
+## **The eval harness + 5MB fictional corpus now live in a sibling repo; gbrain exposes a clean public API they consume.**
+
+BrainBench is gbrain's benchmark harness. 10/12 Cats, 4-adapter scorecard, 418-item fictional corpus, 314 tests. Previously it lived inside this repo. Every `bun install` pulled down the eval tree, `docs/benchmarks/*.md` reports, `pdf-parse` devDep, and auxiliary test fixtures whether or not you ever ran a benchmark. For the 99% of gbrain users who want a knowledge-brain CLI, that's ~5MB of noise.
+
+v0.20 moves BrainBench to [github.com/garrytan/gbrain-evals](https://github.com/garrytan/gbrain-evals). gbrain stays the knowledge-brain CLI + library. `gbrain-evals` depends on gbrain via GitHub URL and consumes it through the public exports map. Same benchmarks, same scorecards, same Cat runners, same 418-item fictional amara-life corpus, just a separate install. Folks who don't care about evals never download them. Folks who do clone one extra repo.
+
+The clean separation also gives gbrain a first real public API surface. `package.json` adds 11 new subpath exports (`gbrain/engine`, `gbrain/pglite-engine`, `gbrain/search/hybrid`, `gbrain/link-extraction`, `gbrain/extract`, and so on) covering every gbrain internal the eval harness reaches into. Third-party tools (not just BrainBench) now have a stable contract to consume. Removing any of these exports is a breaking change going forward.
+
+### What moved where
+
+| Stays in gbrain | Moves to gbrain-evals |
+|-----------------|----------------------|
+| `src/` (CLI, MCP, engines, operations, skills runtime) | `eval/` (runners, adapters, generators, schemas, gold, cli) |
+| `Page.type` enum including `email/slack/calendar-event/note/meeting` (useful for any ingested format, not just evals) | `test/eval/` (314 tests across 14 files) |
+| `inferType()` heuristics for the new directory patterns | `docs/benchmarks/*.md` (all scorecards + regression reports) |
+| Public exports map (11 new subpaths gbrain-evals consumes) | `pdf-parse` devDep (only eval/runner/loaders/pdf.ts used it) |
+| `src/core/` test suite (1696 tests) | `eval:*` scripts (run from gbrain-evals now) |
+
+### What this means for you
+
+If you install gbrain via `git clone + bun install` or via npm/clawhub, you get a smaller, cleaner checkout. No eval corpus. No benchmark reports. No pdf-parse. `bun test` runs only gbrain's own test suite, not eval tests.
+
+If you want to run BrainBench: `git clone https://github.com/garrytan/gbrain-evals && cd gbrain-evals && bun install && bun run eval:run`. gbrain-evals fetches gbrain from GitHub via `"gbrain": "github:garrytan/gbrain#master"` so you always benchmark against the latest source.
+
+If you're a third-party library author importing gbrain internals: the new exports map is now your stable contract. Pin `gbrain/<subpath>` imports against a version, not a file path.
+
+### Itemized changes
+
+**Extracted to [gbrain-evals](https://github.com/garrytan/gbrain-evals):**
+- `eval/` ... schemas, runners, adapters, generators, queries, CLI tools, docs (CONTRIBUTING, RUNBOOK, CREDITS).
+- `test/eval/` ... 14 test files, 314 tests covering schemas, sealed qrels, tool-bridge, agent adapter, judge, recorder, Cat 5/6/8/9/11, amara-life skeleton, adversarial-injections, pdf loader.
+- `docs/benchmarks/` ... all scorecards and regression reports (4-adapter, v0.11 vs v0.12, Minions production/lab, tweet ingestion, knowledge runtime v0.13, BrainBench v1).
+- `pdf-parse` devDep ... only consumed by `eval/runner/loaders/pdf.ts`.
+- `eval:*` package.json scripts ... now live in gbrain-evals's `package.json` and run from there.
+
+**Kept in gbrain (useful beyond evals):**
+- `Page.type` enum extensions in `src/core/types.ts`: `email | slack | calendar-event | note | meeting`. Any user ingesting an inbox dump, Slack export, iCal file, or meeting transcript benefits from first-class types.
+- `inferType()` heuristics in `src/core/markdown.ts` for `/emails/`, `/slack/`, `/cal/`, `/notes/`, `/meetings/` directory patterns.
+- 11 new public `exports` in `package.json`: `./pglite-engine`, `./link-extraction`, `./import-file`, `./transcription`, `./embedding`, `./config`, `./markdown`, `./backoff`, `./search/hybrid`, `./search/expansion`, `./extract`. These form gbrain's public-API contract for downstream consumers.
+
+**Docs synced:**
+- `README.md` ... benchmark references now point at the gbrain-evals repo.
+- `CLAUDE.md` ... BrainBench section replaced with a pointer to gbrain-evals + the list of public exports that consumers depend on.
+- `src/commands/migrations/v0_12_0.ts` ... migration banner text references `github.com/garrytan/gbrain-evals` instead of a local `docs/benchmarks/*.md` path that no longer resolves.
+
+**Tests:** 1717 gbrain tests pass, 0 failures, 174 skipped (E2E requiring `DATABASE_URL`). The full eval suite (314 tests) moves with `gbrain-evals` and runs from there.
+
+### To take advantage of v0.20
+
+For gbrain users:
+1. `gbrain upgrade` ... no action required. The extraction is transparent.
+2. If you previously ran `bun run eval:*` scripts from this repo: those scripts no longer exist here. `git clone https://github.com/garrytan/gbrain-evals && bun install` to get them.
+
+For gbrain-evals consumers:
+1. Clone the sibling repo: `git clone https://github.com/garrytan/gbrain-evals`
+2. `bun install && bun run eval:run`
+3. Follow `gbrain-evals/eval/RUNBOOK.md` for full category runs and scorecard reproduction.
+
 ## [0.19.1] - 2026-04-24
 
 ### Added
diff --git a/CLAUDE.md b/CLAUDE.md
index 830f2af77..8e6bffe26 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -44,7 +44,14 @@ strict behavior when unset.
 - `src/core/embedding.ts` — OpenAI text-embedding-3-large, batch, retry, backoff
 - `src/core/check-resolvable.ts` — Resolver validation: reachability, MECE overlap, DRY checks, structured fix objects. v0.14.1: `CROSS_CUTTING_PATTERNS.conventions` is an array (notability gate accepts both `conventions/quality.md` and `_brain-filing-rules.md`). New `extractDelegationTargets()` parses `> **Convention:**`, `> **Filing rule:**`, and inline backtick references. DRY suppression is proximity-based via `DRY_PROXIMITY_LINES = 40`.
 - `src/core/repo-root.ts` — Shared `findRepoRoot(startDir?)` (v0.16.4): walks up from `startDir` (default `process.cwd()`) looking for `skills/RESOLVER.md`. Zero-dependency module imported by both `doctor.ts` and `check-resolvable.ts`. Parameterized `startDir` makes tests hermetic.
-- `src/commands/check-resolvable.ts` — Standalone CLI wrapper (v0.16.4) over `checkResolvable()`. Exports `parseFlags`, `resolveSkillsDir`, `DEFERRED`, `runCheckResolvable`. Exit rule: **1 on any issue (warnings OR errors)**, stricter than doctor's `ok` flag — honors README:259. Stable JSON envelope `{ok, skillsDir, report, autoFix, deferred, error, message}` — same shape on success and error paths. `--fix` path runs `autoFixDryViolations` BEFORE `checkResolvable` (same ordering as doctor). `deferred[]` array surfaces pending Checks 5 (trigger routing eval) and 6 (brain filing) with issue URLs. `scripts/skillify-check.ts` subprocess-calls `gbrain check-resolvable --json` (cached per process) and fails loud on binary-missing — no silent false-pass.
+- `src/commands/check-resolvable.ts` — Standalone CLI wrapper (v0.16.4) over `checkResolvable()`. Exports `parseFlags`, `resolveSkillsDir`, `DEFERRED`, `runCheckResolvable`. Exit rule: **1 on any issue (warnings OR errors)**, stricter than doctor's `ok` flag — honors README:259. Stable JSON envelope `{ok, skillsDir, report, autoFix, deferred, error, message}` — same shape on success and error paths. `--fix` path runs `autoFixDryViolations` BEFORE `checkResolvable` (same ordering as doctor). `scripts/skillify-check.ts` subprocess-calls `gbrain check-resolvable --json` (cached per process) and fails loud on binary-missing — no silent false-pass. **v0.19:** AGENTS.md workspaces now resolve natively (see `src/core/resolver-filenames.ts`) — gbrain inspects the 107-skill OpenClaw deployment whether the routing file is `RESOLVER.md` or `AGENTS.md`. `DEFERRED[]` is empty — Checks 5 + 6 shipped as real code, not issue URLs.
+- `src/core/resolver-filenames.ts` (v0.19) — central list of accepted routing filenames (`RESOLVER.md`, `AGENTS.md`). Shared by `findRepoRoot`, `check-resolvable`, and skillpack install so every code path walks the same fallback chain.
+- `src/commands/skillify.ts` + `src/core/skillify/{generator,templates}.ts` (v0.19) — `gbrain skillify scaffold <name>` creates all stubs for a new skill in one command: SKILL.md, script, tests, routing-eval.jsonl, resolver entry, filing-rules pointer. `gbrain skillify check <script>` runs the 10-step checklist (LLM evals, routing evals, check-resolvable gate, filing audit) against a candidate skill before it lands.
+- `src/commands/skillify-check.ts` (v0.19) — `gbrain skillpack-check` agent-readable health report. Exit 0/1/2 for CI pipeline gating; JSON for debugging. Wraps `check-resolvable --json`, `doctor --json`, and migration ledger into one payload so agents can decide whether a human action is required.
+- `src/commands/skillpack.ts` + `src/core/skillpack/{bundle,installer}.ts` (v0.19) — `gbrain skillpack install` drops gbrain's curated 25-skill bundle into a host workspace, managed-block style. Never clobbers local edits; tracks a skill manifest so subsequent `install --update` diffs cleanly. Bundle builder (`skillpack/bundle.ts`) packages the set from `skills/` into a versioned payload.
+- `src/core/skill-manifest.ts` (v0.19) — parser for `skill-manifest.json` records. Used by skillpack installer to detect drift between the shipped bundle and the user's local edits, so updates merge instead of overwriting.
+- `src/commands/routing-eval.ts` + `src/core/routing-eval.ts` (v0.19) — `gbrain routing-eval` catches user phrasings that route to the wrong skill. Reads `skills/<name>/routing-eval.jsonl` fixtures (`{intent, expected_skill, ambiguous_with?}`). Structural layer runs in `check-resolvable` by default (zero API cost); `--llm` opts into a Haiku tie-break layer for CI. False positives surface before users hit them.
+- `src/core/filing-audit.ts` + `skills/_brain-filing-rules.json` (v0.19) — Check 6 of `check-resolvable`. Parses new `writes_pages:` / `writes_to:` frontmatter on skills and audits their filing claims against the filing-rules JSON. Warning-only in v0.19, upgrades to error in v0.20.
 - `src/core/dry-fix.ts` — `gbrain doctor --fix` engine. `autoFixDryViolations(fixes, {dryRun})` rewrites inlined rules to `> **Convention:** see [path](path).` callouts via three shape-aware expanders (bullet / blockquote / paragraph). Five guards: working-tree-dirty (`getWorkingTreeStatus()` returns 3-state `'clean' | 'dirty' | 'not_a_repo'`), no-git-backup, inside-code-fence, already-delegated (40-line proximity, consistent with detector), ambiguous-multi-match, block-is-callout. `execFileSync` array args (no shell — no injection surface). EOF newline preserved.
 - `src/core/backoff.ts` — Adaptive load-aware throttling: CPU/memory checks, exponential backoff, active hours multiplier
 - `src/core/fail-improve.ts` — Deterministic-first, LLM-fallback loop with JSONL failure logging and auto-test generation
@@ -113,7 +120,7 @@ strict behavior when unset.
 - `docs/guides/diligence-ingestion.md` — Data room to brain pages pipeline
 - `docs/designs/HOMEBREW_FOR_PERSONAL_AI.md` — 10-star vision for integration system
 - `docs/mcp/` — Per-client setup guides (Claude Desktop, Code, Cowork, Perplexity)
-- `docs/benchmarks/` — Search quality benchmark results (reproducible, fictional data)
+- BrainBench (benchmark suite + corpus): lives in the separate [gbrain-evals](https://github.com/garrytan/gbrain-evals) repo. Not installed alongside gbrain.
 - `skills/_brain-filing-rules.md` — Cross-cutting brain filing rules (referenced by all brain-writing skills)
 - `skills/RESOLVER.md` — Skill routing table (based on the agent-fork AGENTS.md pattern)
 - `skills/conventions/` — Cross-cutting rules (quality, brain-first, model-routing, test-before-bulk, cross-modal)
@@ -144,6 +151,21 @@ strict behavior when unset.
 - `src/commands/report.ts` — Structured report saver (audit trail for maintenance/enrichment)
 - `openclaw.plugin.json` — ClawHub bundle plugin manifest
 
+### BrainBench — in a sibling repo (v0.20+)
+
+BrainBench — the public benchmark for personal-knowledge agent stacks — lives in
+[github.com/garrytan/gbrain-evals](https://github.com/garrytan/gbrain-evals). It
+depends on gbrain as a consumer; gbrain never pulls in the ~5MB eval corpus or
+the pdf-parse dev dep at install time.
+
+gbrain's public API surface (the exports map in `package.json`) is what
+gbrain-evals consumes: `gbrain/engine`, `gbrain/types`, `gbrain/operations`,
+`gbrain/pglite-engine`, `gbrain/link-extraction`, `gbrain/import-file`,
+`gbrain/transcription`, `gbrain/embedding`, `gbrain/config`, `gbrain/markdown`,
+`gbrain/backoff`, `gbrain/search/hybrid`, `gbrain/search/expansion`,
+`gbrain/extract`. Removing any of these is a breaking change for the
+gbrain-evals consumer.
+
 ## Commands
 
 Run `gbrain --help` or `gbrain --tools-json` for full command reference.
@@ -236,7 +258,15 @@ parity), `test/cli.test.ts` (CLI structure), `test/config.test.ts` (config redac
 `test/sync.test.ts` (sync logic + v0.12.3 regression guard asserting top-level `engine.transaction` is not called),
 `test/doctor.test.ts` (doctor command + v0.12.3 assertions that `jsonb_integrity` scans the four v0.12.0 write sites and `markdown_body_completeness` is present),
 `test/utils.test.ts` (shared SQL utilities + `tryParseEmbedding` null-return and single-warn semantics),
-`test/build-llms.test.ts` (llms.txt/llms-full.txt generator: path resolution, idempotence, spec shape, regen-drift guard, content contract, AGENTS.md install-path mirror, size-budget enforcement — 7 cases).
+`test/build-llms.test.ts` (llms.txt/llms-full.txt generator: path resolution, idempotence, spec shape, regen-drift guard, content contract, AGENTS.md install-path mirror, size-budget enforcement — 7 cases),
+`test/check-resolvable-cli.test.ts` (v0.19 CLI wrapper: exit codes, JSON envelope shape, AGENTS.md fallback chain),
+`test/regression-v0_16_4.test.ts` (findRepoRoot regression guard — hermetic startDir parameterization),
+`test/filing-audit.test.ts` (v0.19 Check 6: `writes_pages` / `writes_to` frontmatter, filing-rules JSON validation),
+`test/routing-eval.test.ts` (v0.19 Check 5: fixture parsing, structural routing, ambiguous_with, Haiku tie-break layer),
+`test/skill-manifest.test.ts` (v0.19 skill manifest parser: drift detection, managed-block markers),
+`test/skillify-scaffold.test.ts` (v0.19 `gbrain skillify scaffold` stubs: SKILL.md, script, tests, routing-eval fixtures),
+`test/skillpack-install.test.ts` (v0.19 `gbrain skillpack install` managed-block install / update / no-clobber semantics),
+`test/skillpack-sync-guard.test.ts` (v0.19 sync-guard: bundled skills stay byte-identical to `skills/` source).
 
 E2E tests (`test/e2e/`): Run against real Postgres+pgvector. Require `DATABASE_URL`.
 - `bun run test:e2e` runs Tier 1 (mechanical, all operations, no API keys). Includes 9 dedicated cases for the postgres-engine `addLinksBatch` / `addTimelineEntriesBatch` bind path — postgres-js's `unnest()` binding is structurally different from PGLite's and gets its own coverage.
@@ -245,6 +275,7 @@ E2E tests (`test/e2e/`): Run against real Postgres+pgvector. Require `DATABASE_U
 - `test/e2e/postgres-jsonb.test.ts` — v0.12.2 regression test. Round-trips all 5 JSONB write sites (pages.frontmatter, raw_data.data, ingest_log.pages_updated, files.metadata, page_versions.frontmatter) against real Postgres and asserts `jsonb_typeof='object'` plus `->>'key'` returns the expected scalar. The test that should have caught the original double-encode bug.
 - `test/e2e/jsonb-roundtrip.test.ts` — v0.12.3 companion regression against the 4 doctor-scanned JSONB sites. Assertion-level overlap with `postgres-jsonb.test.ts` is intentional defense-in-depth: if doctor's scan surface ever drifts from the actual write surface, one of these tests catches it.
 - `test/e2e/upgrade.test.ts` runs check-update E2E against real GitHub API (network required)
+- `test/e2e/openclaw-reference-compat.test.ts` (v0.19) — exercises `check-resolvable` + `skillpack install` against a minimal AGENTS.md workspace fixture (`test/fixtures/openclaw-reference-minimal/`), regression guard for the 107-skill OpenClaw deployment shape
 - Tier 2 (`skills.test.ts`) requires OpenClaw + API keys, runs nightly in CI
 - If `.env.testing` doesn't exist in this directory, check sibling worktrees for one:
   `find ../  -maxdepth 2 -name .env.testing -print -quit` and copy it here if found.
@@ -465,7 +496,7 @@ Voice rules:
 
 Source material to pull from:
 - CHANGELOG.md previous entry for prior context
-- `docs/benchmarks/[latest].md` for the headline numbers
+- Latest `gbrain-evals/docs/benchmarks/[latest].md` for headline numbers (sibling repo)
 - Recent commits (`git log <prev-version>..HEAD --oneline`) for what shipped
 - Don't make up numbers. If a metric isn't in a benchmark or production data, don't
   include it. Say "no measurement yet" if asked.
diff --git a/README.md b/README.md
index 8865c80c2..047cd9cf1 100644
--- a/README.md
+++ b/README.md
@@ -4,7 +4,7 @@ Your AI agent is smart but forgetful. GBrain gives it a brain.
 
 Built by the President and CEO of Y Combinator to run his actual AI agents. The production brain powering his OpenClaw and Hermes deployments: **17,888 pages, 4,383 people, 723 companies**, 21 cron jobs running autonomously, built in 12 days. The agent ingests meetings, emails, tweets, voice calls, and original ideas while you sleep. It enriches every person and company it encounters. It fixes its own citations and consolidates memory overnight. You wake up and the brain is smarter than when you went to bed.
 
-The brain wires itself. Every page write extracts entity references and creates typed links (`attended`, `works_at`, `invested_in`, `founded`, `advises`) with zero LLM calls. Hybrid search. Self-wiring knowledge graph. Structured timeline. Backlink-boosted ranking. Ask "who works at Acme AI?" or "what did Bob invest in this quarter?" and get answers vector search alone can't reach. Benchmarked end-to-end: **Recall@5 jumps from 83% to 95%, Precision@5 from 39% to 45%, +30 more correct answers in the agent's top-5 reads** on a 240-page Opus-generated rich-prose corpus. Graph-only F1: **86.6% vs grep's 57.8%** (+28.8 pts). [Full report](docs/benchmarks/2026-04-18-brainbench-v1.md).
+The brain wires itself. Every page write extracts entity references and creates typed links (`attended`, `works_at`, `invested_in`, `founded`, `advises`) with zero LLM calls. Hybrid search. Self-wiring knowledge graph. Structured timeline. Backlink-boosted ranking. Ask "who works at Acme AI?" or "what did Bob invest in this quarter?" and get answers vector search alone can't reach. Benchmarked side-by-side against the category: gbrain lands **P@5 49.1%, R@5 97.9%** on a 240-page Opus-generated rich-prose corpus, beating its own graph-disabled variant by **+31.4 points P@5** and ripgrep-BM25 + vector-only RAG by a similar margin. The graph layer plus v0.12 extract quality together carry the gap. Full BrainBench scorecards + corpus live in the sibling [gbrain-evals](https://github.com/garrytan/gbrain-evals) repo.
 
 GBrain is those patterns, generalized. 28 skills. Install in 30 minutes. Your agent does the work. As Garry's personal agent gets smarter, so does yours.
 
@@ -194,7 +194,7 @@ Here's my personal OpenClaw deployment: one Render container. Supabase Postgres
 
 Under that 19-cron load, sub-agent spawn couldn't clear the 10-second gateway wall. Minions landed it in under a second for zero tokens. **Scaling:** 19,240 posts across 36 months, single bash loop, ~15 min total, $0.00. Sub-agents: ~9 min best case, ~$1.08 in tokens, ~40% spawn failure. **Lab:** durability ∞ (SIGKILL mid-flight, 10/10 rescued), throughput ~10× faster, fan-out ~21× with no failure wall, memory ~400× less.
 
-Full benchmarks: [production](docs/benchmarks/2026-04-18-minions-vs-openclaw-production.md) and [lab](docs/benchmarks/2026-04-18-minions-vs-openclaw-subagents.md).
+Full benchmarks live in [gbrain-evals](https://github.com/garrytan/gbrain-evals/tree/main/docs/benchmarks).
 
 ### The routing rule
 
@@ -438,7 +438,7 @@ gbrain extract links --source db        # wire up the existing 29K pages
 gbrain extract timeline --source db     # extract dated events from markdown timelines
 ```
 
-Then ask graph questions or watch the search ranking improve. Benchmarked: **Recall@5 jumps from 83% to 95%, Precision@5 from 39% to 45%, +30 more correct answers in the agent's top-5 reads** on a 240-page Opus-generated rich-prose corpus. Graph-only F1 hits 86.6% vs grep's 57.8% (+28.8 pts). See [docs/benchmarks/2026-04-18-brainbench-v1.md](docs/benchmarks/2026-04-18-brainbench-v1.md).
+Then ask graph questions or watch the search ranking improve. Benchmarked side-by-side against ripgrep-BM25, vector-only RAG (same embedder), and gbrain-with-graph-disabled: gbrain lands **P@5 49.1%, R@5 97.9%** on a 240-page Opus-generated rich-prose corpus, beating hybrid-nograph by **+31.4 points P@5**. Isolate the contribution: v0.11→v0.12 moved the same gbrain codebase from P@5 22.1% → 49.1% on identical inputs, so typed-link extract quality is load-bearing. Full scorecards + reproducible corpus: [gbrain-evals](https://github.com/garrytan/gbrain-evals).
 
 ## Search
 
@@ -514,7 +514,7 @@ End-to-end on the BrainBench v1 corpus (240 rich-prose pages, before/after PR #1
 
 Plus 5 orthogonal capability checks (identity resolution, temporal queries,
 performance at 10K-page scale, robustness to malformed input, MCP operation
-contract). All pass. [Full report.](docs/benchmarks/2026-04-18-brainbench-v1.md)
+contract). All pass. Full report: [gbrain-evals](https://github.com/garrytan/gbrain-evals).
 
 The point: each technique handles a class of inputs the others miss. Vector
 search misses exact slug refs; keyword catches them. Keyword misses conceptual
@@ -676,7 +676,7 @@ The skills in this repo are those patterns, generalized. What took 11 days to bu
 - [CHANGELOG.md](CHANGELOG.md) ... Version history
 
 **Benchmarks:**
-- [BrainBench v1 (PR #188)](docs/benchmarks/2026-04-18-brainbench-v1.md) ... single comprehensive before/after report on a 240-page Opus-generated corpus. 7 categories: relational queries, identity resolution, temporal queries, performance, robustness, MCP contract.
+- [gbrain-evals](https://github.com/garrytan/gbrain-evals) ... BrainBench, the sibling repo that holds the eval harness, corpus, scorecards, and 4-adapter comparisons. Depends on gbrain; not installed alongside gbrain.
 
 ## Contributing
 
diff --git a/TODOS.md b/TODOS.md
index f47442408..3e2608e99 100644
--- a/TODOS.md
+++ b/TODOS.md
@@ -11,44 +11,45 @@ Both checks shipped as real implementations, not just filed issues:
 
 `DEFERRED[]` in `src/commands/check-resolvable.ts` is now empty — v0.19 shipped both deferred checks as working code paths, not as issue URLs. The export stays in place for future deferred checks.
 
-## P1 (BrainBench v1.1 — categories deferred from PR #188)
+### ~~BrainBench Cats 5/6/8/9/11 — shipped to sibling repo~~
+**Completed:** v0.20.0 (2026-04-23)
 
-### BrainBench Cat 5: Source Attribution / Provenance
-**What:** Eval that gbrain correctly cites the right page when claiming fact F, and resolves source-conflict cases (3 sources disagree on $5M raise — which wins?). 200 queries across citation/provenance/conflict sub-categories on a 300-entity dataset with deliberately-conflicting sources.
+All five previously-deferred BrainBench categories shipped as working runners
+in the sibling repo [github.com/garrytan/gbrain-evals](https://github.com/garrytan/gbrain-evals):
 
-**Why deferred from PR #188:** Needs ~$100-200 of Opus tokens to generate the conflict-graph dataset. v1 scope was procedural-only.
+- **Cat 5 Provenance** — `eval/runner/cat5-provenance.ts` with dedicated `classify_claim` tool (3-way label: `supported | unsupported | over-generalized`)
+- **Cat 6 Prose-scale auto-link precision** — `eval/runner/cat6-prose-scale.ts` (baseline-only) + `eval/runner/adversarial-injections.ts` (6 injection kinds)
+- **Cat 8 Skill Compliance** — `eval/runner/cat8-skill-compliance.ts` (brain-first / back-link / citation-format / tier-escalation, deterministic from tool-bridge trace)
+- **Cat 9 End-to-End Workflows** — `eval/runner/cat9-workflows.ts` (rubric-graded)
+- **Cat 11 Multi-modal Ingestion** — `eval/runner/cat11-multimodal.ts` (PDF/audio/HTML)
 
-**Threshold:** citation_recall > 90%, citation_precision > 85%, conflict_resolution > 70%.
+Plus supporting infrastructure: agent adapter (Sonnet + 12 read + 3 dry_run tools),
+structured-evidence Haiku judge contract, PublicPage/PublicQuery sealed qrels,
+6-artifact flight-recorder, 6 portable JSON schemas for v1→v2 driver swap.
 
-**Depends on:** Identity Resolution (Cat 3) shipped — uses same world generator pattern.
+Scope pivot: originally planned for in-tree v1.1 delta; mid-PR pivoted to extract
+the entire eval harness so gbrain users don't download the ~5MB corpus at install
+time. BrainBench is now a public sibling benchmark; gbrain ships clean.
 
-### BrainBench Cat 6: Auto-link Precision under Prose (at scale)
-**What:** Cat 10 (Robustness/Adversarial) covered code-fence leak and false-positive substrings on 22 hand-crafted cases. v1.1 extends this to 500+ prose-heavy pages with realistic narrative noise. Tests link precision in the wild, not just edge cases.
+### ~~v0.10.5: inferLinkType residuals (works_at, advises)~~
+**Completed:** v0.20.0 (2026-04-23)
 
-**Why deferred from PR #188:** Needs prose-heavy generated corpus (~$100-150 Opus). Existing 22-case eval already caught + fixed the code-fence leak bug.
+`src/core/link-extraction.ts` — WORKS_AT_RE and ADVISES_RE expanded with
+rank-prefixed engineer patterns ("senior/staff/principal/lead engineer at"),
+discipline-prefixed ("backend/frontend/ML/security engineer at"), broader role
+verbs ("manages engineering at", "running product at", "heads up X at"),
+possessive time ("his/her/their time at"), role-noun forms ("tenure as",
+"stint as", "role at"), advisory capacity phrasings, "as an advisor" forms,
+and qualifier-specific advisors. New EMPLOYEE_ROLE_RE prior fires for
+self-identified employees at the page level, biasing outbound company refs
+toward works_at when per-edge verbs are absent. Precedence: investor > advisor
+> employee. Existing tests in `test/link-extraction.test.ts` cover the new
+patterns.
 
-**Threshold:** link_precision > 95% on prose, type_accuracy > 80% on varied phrasing.
+## P1 (BrainBench v1.1 — remaining categories)
 
-### BrainBench Cat 8: Skill Behavior Compliance
-**What:** Replays 100 inbound signals through a real LLM agent loop with gbrain skills loaded. Measures: brain-first lookup compliance, back-link iron-law adherence, citation format compliance, tier escalation correctness.
-
-**Why deferred:** Needs real LLM API loop (~$2K total — most expensive single category).
-
-**Threshold:** brain_first_compliance > 95%, back_link_compliance > 90%, citation_format > 95%.
-
-### BrainBench Cat 9: End-to-End Workflows
-**What:** 50 end-to-end scenarios across meeting ingestion, email-to-brain, daily-task-prep, briefing generation, sync cycle. Rubric-graded (10-15 criteria each).
-
-**Why deferred:** Needs LLM agent loop (~$1K). Plus 50 hand-built rubrics.
-
-**Threshold:** 80% scenario pass rate per workflow.
-
-### BrainBench Cat 11: Multi-modal Ingestion
-**What:** PDF/image/audio/video ingestion accuracy. 50 PDFs, 30 images, 20 audio files, 10 videos, 30 HTML pages. Per-modality recall and fidelity metrics.
-
-**Why deferred:** Needs licensed real datasets (Common Voice for audio etc.). Dataset curation is the bulk of the work.
-
-**Threshold:** PDF text fidelity > 95% (text-based) / > 80% (scanned), audio WER < 15%, entity_recall > 80% post-ingestion.
+Cats 5/6/8/9/11 shipped to the sibling repo in v0.20.0 — see the Completed
+section above. One remaining scope item:
 
 ### BrainBench Cat 1+2 at full scale
 **What:** Existing benchmark-search-quality.ts (29 pages, 20 queries) and benchmark-graph-quality.ts (80 pages, 5 queries) currently pass at small scale. v1.1 extends both to 2-3K rich-prose pages generated via Opus to surface scale-dependent failures (tied keyword clusters, hub-node fan-out, prose-noise extraction precision).
@@ -68,24 +69,6 @@ company refs only). Per-type after fix: invested_in 91.7% (was 0%),
 mentions 100%, attended 100%. works_at 58% and advises 41% are next
 iteration's residuals.
 
-### v0.10.5: inferLinkType residuals (works_at, advises)
-**What:** After the v0.10.4 fix, two link types still under-perform on rich
-prose. Drive these to >85% type accuracy in next iteration.
-
-**works_at: 58% type accuracy.** Engineer/employee pages use varied phrasings
-the regex doesn't catch ("spent some time at", "joined the team", narrative
-"is currently at" without a verb). Approach: extend WORKS_AT_RE; consider
-employee-role page prior similar to partner prior.
-
-**advises: 41% type accuracy.** Advisor pages often describe board roles
-without using the word "advisor" explicitly ("on Beta Health's board",
-"joined Beta as a board member"). The v0.10.4 fix tightened ADVISES_RE to
-require "advisor" rooting to avoid false positives from investors. Need
-a tighter signal that distinguishes "advisor on board" from "investor on
-board" — likely an advisor-role page prior plus verb-pattern combinations.
-
-**Threshold:** Cat 2 rich-prose type accuracy > 92% (currently 88.5%).
-
 ### v0.10.4: gbrain alias resolution feature (driven by Cat 3)
 **What:** Add an alias table to gbrain so "Sarah Chen" / "S. Chen" / "@schen" / "sarah.chen@example.com" resolve to one canonical entity. Schema: `aliases (id, slug, alias_text)` with a unique index. Search blends alias matches into hybrid scoring.
 
diff --git a/VERSION b/VERSION
index 41915c799..5a03fb737 100644
--- a/VERSION
+++ b/VERSION
@@ -1 +1 @@
-0.19.1
+0.20.0
diff --git a/docs/benchmarks/2026-04-18-brainbench-v1.md b/docs/benchmarks/2026-04-18-brainbench-v1.md
deleted file mode 100644
index 7458c32dc..000000000
--- a/docs/benchmarks/2026-04-18-brainbench-v1.md
+++ /dev/null
@@ -1,286 +0,0 @@
-# BrainBench v1 — 2026-04-18
-
-**Branch:** `garrytan/link-timeline-extract`
-**PR:** #188
-**Engine:** PGLite (in-memory)
-**Reproducibility:** `bun run eval/runner/all.ts` — no API keys, no network, ~3 min
-
-## TL;DR
-
-PR #188 ships a self-wiring knowledge graph layer for gbrain (auto-link on
-every page write, typed extraction, traversal queries, backlink-boosted search).
-This benchmark measures the actual end-to-end value vs gbrain pre-PR-#188 on a
-240-page rich-prose corpus generated by Claude Opus.
-
-**Every headline metric goes UP. No category goes down.**
-
-| Metric              | BEFORE PR #188 | AFTER PR #188 | Δ            |
-|---------------------|----------------|---------------|--------------|
-| **Precision@5**     | 39.2%          | **44.7%**     | **+5.4 pts** |
-| **Recall@5**        | 83.1%          | **94.6%**     | **+11.5 pts**|
-| Correct in top-5    | 217            | 247           | **+30**      |
-
-Plus seven categories of orthogonal capability checks (identity resolution,
-temporal queries, performance, robustness, MCP contract) all passing.
-
-## What this benchmark proves
-
-BrainBench v1 evaluates gbrain end-to-end across capability domains the existing
-test suite doesn't cover at scale. Headline is a single before/after comparison:
-**pre-PR-#188 (no graph layer)** vs **the full v0.10.3 + v0.10.4 stack**, run on
-the same 240-page corpus with the same relational queries.
-
-Why before/after instead of just "after numbers": because gbrain pre-PR-#188 was
-already a working brain — keyword search, hybrid retrieval, structured timeline
-ops. The graph layer is an additive change. The right question is "did it
-actually make the brain better at relational questions?" not "is it good in
-isolation."
-
-## The corpus
-
-240 rich-prose pages generated by Claude Opus 4.7:
-- 80 people (40 founders, 20 partners, 10 engineers, 10 advisors)
-- 80 companies (60 startups, 15 VCs, 5 acquirers)
-- 50 meetings (15 demo days, 25 1:1s, 10 board meetings)
-- 30 concepts (frameworks, theses, hot spaces)
-
-Each page is multi-paragraph narrative prose with realistic noise:
-- Varied phrasings (founders described 6 different ways, investors 8 different ways)
-- Natural typos ~1-2% of words ("intrest", "comercial", "differnt")
-- Cross-references via `[Name](slug)` markdown links AND bare slug references
-- Multi-year timelines spanning 2021-2026
-- Multiple personas (terse note-taker, prose-heavy journaler, voice-to-text dump)
-
-Generation cost: ~$15 of Opus tokens, one-time, cached to `eval/data/world-v1/`
-and committed to the repo. Subsequent runs read the cache.
-
-This is intentionally messier than templated benchmarks. The point is to surface
-behavior under realistic load, not to confirm the algorithm works on clean inputs.
-
-## Headline: relational queries on the rich corpus
-
-196 relational queries derived from the world facts:
-- "Who attended `Demo Day W30`?" (60 queries)
-- "Who works at `Acme`?" (60 queries)
-- "Who invested in `Beta Health`?" (45 queries)
-- "Who advises `Cipher Labs`?" (31 queries)
-
-Configurations compared:
-- **BEFORE PR #188:** vanilla v0.10.0 — no auto-link, no `extract --source db`,
-  no `traversePaths`. Agent answers relational questions by grepping the corpus
-  (the realistic fallback for a pre-graph brain).
-- **AFTER PR #188:** full graph layer. Agent uses `gbrain graph-query` first
-  (high-precision typed traversal), grep fallback when graph returns nothing.
-
-### Top-K (what agents actually read)
-
-Agents read ranked top-K results, not full sets. AFTER ranks graph hits FIRST
-(high precision), then fills with grep results.
-
-| Metric              | BEFORE | AFTER  | Δ             |
-|---------------------|--------|--------|---------------|
-| **Precision@5**     | 39.2%  | 44.7%  | **+5.4 pts**  |
-| **Recall@5**        | 83.1%  | 94.6%  | **+11.5 pts** |
-| Correct in top-5    | 217    | 247    | **+30**       |
-
-Recall@5 jumps 11.5 points because graph hits are exact-typed answers placed
-at the top of results — agents find what they need in their first reads
-instead of digging through grep noise.
-
-### Set-based metrics + graph-only ablation
-
-| Metric              | BEFORE (grep) | AFTER (hybrid) | Graph-only (ablation) |
-|---------------------|---------------|----------------|------------------------|
-| **F1 score**        | 57.8%         | 57.8%          | **86.6%**              |
-| Set precision       | 40.8%         | 40.8%          | **81.0%**              |
-| Set recall          | 98.9%         | 98.9%          | 93.1%                  |
-| Total returned      | 632           | 632            | 300 (-53%)             |
-| Correct returned    | 258           | 258            | 243                    |
-
-AFTER (hybrid) matches BEFORE on full-set metrics because graph hits are a
-subset of grep hits — taking the union doesn't add or remove anything from the
-bag. **What changes is which results appear FIRST.** Top-K captures that;
-raw set recall doesn't.
-
-The **graph-only** column is the most important number in the report. It shows
-where the graph alone is heading: **86.6% F1 vs grep's 57.8% (+28.8 pts)**.
-Almost twice the precision (81% vs 41%) at 94% of the recall, with HALF the
-results to read.
-
-### Per-link-type breakdown
-
-| Link type   | Expected | Graph found / returned | Recall | Precision |
-|-------------|----------|------------------------|--------|-----------|
-| attended    | 134      | 131 / 134              | 97.8%  | 97.8%     |
-| works_at    | 50       | 50 / 79                | 100.0% | 63.3%     |
-| invested_in | 60       | 50 / 56                | 83.3%  | 89.3%     |
-| advises     | 17       | 12 / 31                | 70.6%  | 38.7%     |
-
-Where the graph wins biggest: **incoming relationship queries on companies**.
-"Who works at Acme?" — grep returns every page mentioning Acme (founders,
-investors, advisors, concept pages, other companies that mention it). Graph
-returns just employees with the typed `works_at` link.
-
-## How we got here: bugs surfaced, fixes shipped
-
-The benchmark wasn't passive — it caught real bugs in the same PR that ships
-the graph layer. Each fix landed in a labeled commit:
-
-### Bug 1: Code fence leak in `extractPageLinks`
-
-**Found:** Category 10 (Robustness) — adversarial test cases included pages with
-slug-like strings inside ` ``` ` code blocks. Extraction was treating them as
-real entity references.
-
-**Fix:** `stripCodeBlocks()` helper preserves byte offsets but blanks out
-fenced and inline code before regex matching. Code fence leak rate now 0%.
-
-### Bug 2: `add_timeline_entry` accepted year 99999
-
-**Found:** Category 12 (MCP Contract) — boundary input fuzzing.
-
-**Fix:** Strict YYYY-MM-DD regex with year clamped 1900-2199, round-trip parse
-to catch e.g. Feb 30. Rejects with clear error message.
-
-### Bug 3: `inferLinkType` mis-classified investments as `mentions`
-
-**Found:** Rich-prose corpus showed `invested_in` had **0% type accuracy** —
-60/60 found links classified as `mentions`. Templated tests didn't surface this
-because the templated prose used "invested in" verbatim while LLM prose uses
-"led the Series A", "early investor", "portfolio includes", etc.
-
-**Fix:** Five-part patch:
-1. `INVESTED_RE` extended with narrative verbs LLMs actually use
-2. `ADVISES_RE` tightened to require explicit advisor rooting (not generic "board")
-3. Context window 80→240 chars (catches verbs at sentence distance)
-4. Person-page role prior — partner-bio language → `invested_in` for company refs
-5. Cascade reorder — `invested_in` checked before `advises`
-
-Type accuracy: **70.7% → 88.5% (+18 pts)**. invested_in: **0% → 91.7%**.
-
-### Bug 4: Founder bios mis-classified as `invested_in`
-
-**Found:** Diagnostic on rich corpus showed founder pages like "Carol Wilson is
-the founder of [Anchor]" were getting `invested_in` (because the role prior
-fired and `FOUNDED_RE` only matched the verb form "founded", missing the noun
-form "founder of").
-
-**Fix:** Extended `FOUNDED_RE` with "founder of", "founders include", "the
-founder", etc. Carol's link now correctly types as `founded`. Combined with
-relaxing the "who works at X?" query to accept `works_at` OR `founded` (founders
-are employees by definition), this drove the recall jump from 53.8% → 93.1%.
-
-## Other categories (orthogonal capability checks)
-
-Five additional categories run as part of `bun run eval/runner/all.ts`. All pass.
-
-### Category 3: Identity Resolution
-
-Tests whether gbrain can resolve aliases ("Sarah Chen", "S. Chen", "@schen",
-"sarah.chen@example.com") to one canonical entity. 100 entities × 8 alias types
-= 800 queries.
-
-| Alias category | Recall (top-10) |
-|----------------|-----------------|
-| Documented (in canonical body)     | 100.0% |
-| Undocumented (initials, typos)     | 31.0%  |
-
-Honest baseline: gbrain has no alias table today. Documented aliases work via
-keyword search. Undocumented aliases need v0.10.4 alias-table feature
-(documented in TODOS.md).
-
-### Category 4: Temporal Queries
-
-50 entities × 10-20 dated events spanning 5 years. Tests point queries, range
-queries, recency, and as-of queries.
-
-| Sub-category    | Recall | Precision |
-|-----------------|--------|-----------|
-| Point           | 100%   | 100%      |
-| Range           | 100%   | 100%      |
-| Recency (top-3) | 100%   | —         |
-| As-of           | 100%   | —         |
-
-Structured `timeline_entries` table answers all four query types correctly via
-manual filter+sort logic. Note: there's no native `getStateAtTime` op — the
-as-of queries were resolved by the agent in app code. Native op deferred to v0.10.5.
-
-### Category 7: Performance / Latency
-
-Procedural data at 1K and 10K page scales on PGLite (in-memory). All read ops
-sub-millisecond. Bulk import at 5,800 pages/sec.
-
-| Op                 | 1K P50  | 1K P95  | 10K P50 | 10K P95  |
-|--------------------|---------|---------|---------|----------|
-| get_page           | 0.08ms  | 0.12ms  | 0.08ms  | 0.15ms   |
-| search_keyword     | 0.19ms  | 0.52ms  | 0.20ms  | 0.59ms   |
-| traverse_paths d=2 | 10.1ms  | 12.6ms  | 91.4ms  | 176.4ms  |
-| putPage_single     | 0.12ms  | 0.20ms  | 0.12ms  | 0.42ms   |
-
-Bulk throughput: import 5,848 pages/sec, addLink 8,752 links/sec at 10K scale.
-P95 search latency well under the 200ms threshold.
-
-### Category 10: Robustness / Adversarial
-
-22 hand-crafted edge cases × 6 ops each = 133 attempts. Tests empty pages,
-100K-character pages, CJK/Arabic/Cyrillic/emoji, code fences, false-positive
-substrings, malformed timeline, deeply nested markdown, slugs with edge characters.
-
-**Result: 133/133 ops succeeded, 0 crashes, 0 silent corruption.**
-
-### Category 12: MCP Operation Contract
-
-50 contract tests across trust boundary (local vs remote), input validation
-(slug format, date format), SQL injection resistance, resource exhaustion,
-depth caps. 30 operations × 5 input variants.
-
-**Result: 50/50 pass.** Verifies the v0.10.3 security hardening (depth caps,
-remote auto-link disable, file_upload path confinement, parameterized queries).
-
-## Reproducibility
-
-```bash
-bun run eval/runner/all.ts
-```
-
-In-memory PGLite, no API keys, no network. ~3 minutes wall time. Same numbers
-every run (within deterministic-seed tolerance).
-
-To regenerate the rich-prose corpus from scratch (~$15 Opus spend):
-
-```bash
-bun eval/generators/gen.ts --max 240 --concurrency 6
-```
-
-Generated outputs are cached in `eval/data/world-v1/` and committed to the repo,
-so the regen pass is one-time. Subsequent runs use the cache.
-
-## What this benchmark deliberately doesn't test (BrainBench v1.1, see TODOS.md)
-
-- **Cat 5: Source attribution / provenance** — needs ~$200-300 Opus for a
-  conflict-graph corpus
-- **Cat 6: Auto-link precision under prose at scale** — needs 5K+ adversarial
-  prose pages
-- **Cat 8: Skill behavior compliance** — needs LLM agent loop (~$2K to run)
-- **Cat 9: End-to-end workflows** — needs LLM agent loop (~$1K)
-- **Cat 11: Multi-modal ingestion** — needs licensed real datasets
-
-These five are tracked in `TODOS.md` with budget estimates and depend-on chains.
-
-## Methodology notes
-
-- **Synthetic data, not private brain.** All 240 pages are fictional. Generated
-  by Opus from procedural skeletons in `eval/generators/world.ts`. Reproducibility
-  matters more than realism for a benchmark you can publish.
-- **Two configurations, one corpus.** BEFORE and AFTER run against identical
-  data. The only diff is the codepath (whether the agent has the graph layer
-  available). No corpus tuning per configuration.
-- **No cherry-picking.** Queries are derived programmatically from world facts —
-  every entity that has facts produces queries. No hand-selected "easy wins."
-- **Honest about limitations.** The 5.8pt set-recall gap (graph 93.1% vs grep
-  98.9%) comes from Opus paraphrasing names without markdown links ("Mark Thomas
-  was there" instead of `[Mark Thomas](slug)`). Closing this needs corpus-aware
-  NER, deferred to v0.10.5.
-- **Single-shot benchmarks are fragile** — but every run is reproducible and
-  this is a checkpoint, not the final measure. v1.1 will add the LLM-agent-loop
-  categories that capture more of the realistic agent workflow.
diff --git a/docs/benchmarks/2026-04-18-minions-vs-openclaw-production.md b/docs/benchmarks/2026-04-18-minions-vs-openclaw-production.md
deleted file mode 100644
index 9dfca4ac3..000000000
--- a/docs/benchmarks/2026-04-18-minions-vs-openclaw-production.md
+++ /dev/null
@@ -1,126 +0,0 @@
-# Production Benchmark: Minions vs OpenClaw Sub-agents (Real Deployment)
-
-**Date:** 2026-04-18
-**Environment:** Garry's OpenClaw on Render (ephemeral container, Supabase Postgres)
-**GBrain:** v0.11.0 (minions-jobs branch)
-**OpenClaw:** 2026.4.10
-**Brain:** 45,798 pages, 98K chunks, 25K links, 79K timeline entries
-**Task:** Pull and ingest one month of social posts from an external API into the brain
-
-## Context
-
-This is a **production benchmark**, not a lab test. The existing lab benchmark
-([2026-04-18-minions-vs-openclaw-subagents.md](2026-04-18-minions-vs-openclaw-subagents.md))
-uses trivial prompts on localhost Postgres. This benchmark uses a real 45K-page
-brain on Supabase, pulling real social posts from an external API, and writing
-real brain pages.
-
-## The Task
-
-Pull a month (May 2020) of my social posts from an external API, parse them
-into a structured brain page with frontmatter, engagement metrics, and
-links, commit to the brain repo, and submit a sync job to gbrain.
-
-## Method 1: Minions (deterministic pipeline)
-
-```bash
-# 1. Pull posts from the external API (curl → JSON)
-curl -s -H "Authorization: Bearer $API_BEARER_TOKEN" \
-  "$SOCIAL_API_URL?from=my_account&start=2020-05-01&end=2020-06-01" \
-  > /tmp/bench-posts.json
-
-# 2. Parse + write brain page (python)
-python3 parse_and_write.py
-
-# 3. Git commit
-cd /data/brain && git add media/social/2020-05.md && git commit -m "archive: 2020-05"
-
-# 4. Submit sync to Minions
-gbrain jobs submit sync --params '{"repo":"/data/brain","noPull":true}'
-```
-
-**Result: 753ms total.** 99 posts pulled, page written, committed, sync job queued.
-
-Breakdown:
-- External API call: ~300ms
-- Python parse + write: ~50ms
-- Git commit: ~100ms
-- gbrain jobs submit: ~300ms
-
-Cost: $0.00 (no LLM tokens)
-
-## Method 2: OpenClaw Sub-agent (sessions_spawn)
-
-```javascript
-sessions_spawn({
-  task: "Pull my social posts for June 2020 and save as a brain page...",
-  model: "anthropic/claude-sonnet-4-20250514",
-  mode: "run",
-  runTimeoutSeconds: 120
-})
-```
-
-**Result: GATEWAY TIMEOUT (>10,000ms).** The sub-agent could not even spawn
-within the 10-second gateway timeout. On a production Render container running
-a 45K-page brain with 19 active cron jobs, the gateway is under enough load
-that sub-agent spawning is unreliable.
-
-When sub-agents DO successfully spawn (off-peak), the expected path is:
-1. Gateway receives spawn request (~500ms)
-2. Create session, load context (~2-3s) — AGENTS.md, SOUL.md, skills, memory
-3. Model reads task, plans approach (~2-3s)
-4. Model calls `exec` tool for curl (~1s)
-5. Model calls `exec` tool for python (~1s)
-6. Model calls `exec` tool for git (~1s)
-7. Model reports result (~1s)
-
-**Estimated: 10-15s + ~$0.03 in tokens per invocation**
-
-## Comparison
-
-| Metric | Minions | Sub-agent |
-|--------|---------|-----------|
-| **Wall time** | **753ms** | **>10,000ms** (gateway timeout) |
-| **Token cost** | $0.00 | ~$0.03 per run |
-| **Success rate** | 100% | 0% (timeout on first attempt) |
-| **Survives restart** | Yes (Postgres) | No (dies with process) |
-| **Progress tracking** | `gbrain jobs get <id>` | poll sessions_list |
-| **Auto-retry** | 3 attempts, exponential backoff | manual re-spawn |
-| **Concurrency** | FOR UPDATE SKIP LOCKED | hope-based maxConcurrent |
-| **Steerable** | inbox messages | fire and forget |
-| **Results persisted** | job record | lost on compaction |
-| **Memory** | ~2MB per in-flight job | ~80MB per spawned session |
-
-## The Scaling Story
-
-We pulled 19,240 posts across 36 months (2021-2023) using the Minions
-approach in a single bash loop. Total time: ~15 minutes. Cost: $0.00 in
-LLM tokens.
-
-The same task via sub-agents would require 36 spawns × ~$0.03 = ~$1.08
-in tokens, take 36 × 15s = 9 minutes best-case, and fail on ~40% of
-spawns under load (per the fan-out benchmark).
-
-At scale (100+ months of backfill, or 1000+ batch enrichment jobs),
-Minions is the only viable path. Sub-agents hit the gateway timeout wall,
-burn tokens on deterministic work, and provide no durability.
-
-## When Sub-agents Still Win
-
-Sub-agents are correct for **judgment work**:
-- Email triage (LLM decides priority, drafts reply)
-- Social radar (LLM assesses severity, decides to alert)
-- Meeting prep (LLM synthesizes brain pages into briefing)
-- Cold email research (LLM decides notability)
-
-These tasks require an LLM to make decisions. Minions can't do that —
-its handlers are code, not models. The routing rule:
-
-> **Deterministic** (same input → same steps → same output) → **Minions**
-> **Judgment** (input requires assessment/decision) → **Sub-agents**
-
-## One-Line Summary
-
-Minions completed a production post-ingest pipeline in 753ms for $0.
-Sub-agents couldn't even spawn. For deterministic brain-write work,
-Minions is not incrementally better — it's categorically different.
diff --git a/docs/benchmarks/2026-04-18-minions-vs-openclaw-subagents.md b/docs/benchmarks/2026-04-18-minions-vs-openclaw-subagents.md
deleted file mode 100644
index 408d0fd18..000000000
--- a/docs/benchmarks/2026-04-18-minions-vs-openclaw-subagents.md
+++ /dev/null
@@ -1,203 +0,0 @@
-# Minions vs OpenClaw Subagents Benchmark
-
-**Date:** 2026-04-18
-**Branch:** garrytan/minions-jobs
-**Suite:** `test/e2e/bench-vs-openclaw/`
-**Minions:** v0.11.0 (PR #130)
-**OpenClaw:** 2026.4.10 (44e5b62)
-**Model:** anthropic/claude-haiku-4-5
-
-## Why this benchmark exists
-
-Minions is GBrain's new background job queue, pitched as a durable, cheap
-substitute for spawning OpenClaw subagents via `openclaw agent --local`.
-"Durable" and "cheap" are easy to claim and hard to prove. So we put
-numbers on four specific claims a Minions user would actually care about:
-
-1. **Durability** — when the orchestrator crashes mid-dispatch, does the
-   in-flight work survive?
-2. **Throughput** — how much wall-clock overhead does each system add on
-   top of the underlying LLM call?
-3. **Fan-out** — parent dispatches 10 children in parallel. How fast and
-   how reliable is each side?
-4. **Memory** — what does it cost to keep 10 subagents in flight at once?
-
-Methodology: both sides call the **same** LLM
-(`anthropic/claude-haiku-4-5`) with the **same** trivial prompt
-(`"Reply with just: OK. No other text."`). The delta is the
-queue+dispatch+process-cost on top of identical LLM work.
-
-## Honest caveats up front
-
-- **We do NOT benchmark OpenClaw's gateway multi-agent fan-out.** That
-  requires a custom WebSocket client + an LLM-backed parent agent, ~5×
-  the complexity of this harness. We benchmark `openclaw agent --local`
-  (embedded mode) because that's what users actually script against
-  today when they want "run an agent and get a reply back."
-- **All numbers are point measurements on Garry's laptop** (macOS, Apple
-  Silicon, local Postgres 16 + pgvector in Docker). Not a cluster
-  benchmark. Not an adversarial load test. Reproducible via the files
-  in `test/e2e/bench-vs-openclaw/`.
-- **OpenClaw `--local` is a fire-and-forget process.** If you SIGKILL
-  it mid-dispatch, the reply is gone. This isn't a bug, it's the design.
-  What we're measuring is how much that design choice costs users who
-  need durability.
-- **Small sample sizes** (10 jobs × 3 runs for fan-out, 20 serial for
-  throughput, 10 in-flight for memory). Enough to show order-of-magnitude
-  deltas, not enough to prove tight tails.
-
-## Results
-
-### 1. Durability (SIGKILL mid-flight, 10 jobs)
-
-| System | Delivered | Wall time | p50 per job | p95 per job |
-|--------|-----------|-----------|-------------|-------------|
-| **Minions** | **10 / 10** | 458ms total | 257ms | 410ms |
-| OpenClaw `--local` | **0 / 10** | 22989ms (all SIGKILLed at 500ms) | n/a | n/a |
-
-Setup: Minions side seeds 10 jobs in state `active` with an expired
-`lock_until` (exactly the state a SIGKILLed worker leaves behind). A
-rescue worker starts. It picks up all 10 via `handleStalled` and
-completes them.
-
-OpenClaw side spawns 10 `openclaw agent --local` processes in parallel
-and SIGKILLs each at 500ms. Zero of them managed to emit any output
-before being killed.
-
-**The number that matters: Minions rescued 10 out of 10 stranded
-jobs in under half a second.** OpenClaw has no persistence layer, so
-anything in flight when the process dies is lost. Users can retry by
-re-running the prompt, but the context is gone — they're starting over.
-
-Source: `test/e2e/bench-vs-openclaw/durability.bench.ts`
-
-### 2. Throughput (20 serial dispatches, same LLM call)
-
-| System | p50 | p95 | p99 | Mean | Min | Max | Success |
-|--------|-----|-----|-----|------|-----|-----|---------|
-| **Minions** | **778ms** | **1931ms** | **1931ms** | **911ms** | 639ms | 1931ms | 20/20 |
-| OpenClaw `--local` | 8086ms | 10094ms | 10094ms | 8335ms | 7405ms | 10094ms | 20/20 |
-| **Ratio** | **10.4×** | **5.2×** | **5.2×** | **9.2×** | 11.6× | 5.2× | — |
-
-Setup: both sides call claude-haiku-4-5 with the same prompt. Minions
-goes through `queue.add` → worker claims → handler calls Anthropic SDK
-directly. OpenClaw spawns a fresh `openclaw agent --local` process per
-dispatch.
-
-The ~7 seconds of overhead per OC dispatch isn't the LLM. It's the
-process boot: loading the agent runtime, auth, plugins, MCP servers.
-Every dispatch pays that cost again. The Minions worker stays warm, so
-the overhead is `add` + `claim` + returning the result — roughly 100ms
-on top of the LLM latency itself.
-
-Source: `test/e2e/bench-vs-openclaw/throughput.bench.ts`
-
-### 3. Fan-out (3 runs × 10 children in parallel)
-
-| System | Completed | Mean wall time | Runs (ok/N) | Wall times (ms) |
-|--------|-----------|----------------|-------------|-----------------|
-| **Minions** (concurrency=10) | **30 / 30** | **1090ms** | 10/10, 10/10, 10/10 | 890, 1135, 1245 |
-| OpenClaw (10 parallel spawns) | 17 / 30 | 22598ms | 6/10, 5/10, 6/10 | 22204, 22505, 23084 |
-| **Ratio (wall time)** | — | **~21×** | — | — |
-
-Setup: parent dispatches 10 children concurrently, waits for all.
-Minions uses one worker process with `concurrency=10`. OpenClaw scripts
-10 parallel `openclaw agent --local` spawns — what a user would do today
-without Minions.
-
-Two findings, not one:
-
-1. **Wall time: Minions completes 10 in ~1 second. OC parallel spawn
-   takes ~22 seconds.** The gap scales with the warmup cost: one warm
-   worker amortizes, 10 cold processes pay the bill 10 times.
-2. **OC parallel spawn fails 43% of the time at 10-wide.** Error
-   samples show a mix of LLM rate-limit hits and spawn saturation. We
-   didn't tune this. That's the point — a user who tries to fan out with
-   `--local` without a queue runs into this with no obvious remediation.
-
-Source: `test/e2e/bench-vs-openclaw/fanout.bench.ts`
-
-### 4. Memory (10 in-flight subagents)
-
-| System | Baseline RSS | Peak with 10 in flight | Delta | Processes |
-|--------|--------------|------------------------|-------|-----------|
-| **Minions** | 84 MB | **86 MB** | **+2 MB** | 1 |
-| OpenClaw | n/a | 814 MB (summed across 10) | — | 10 |
-| **Ratio** | — | **~407×** | — | — |
-
-Setup: both sides keep 10 subagents in flight simultaneously. Minions
-side uses one worker with concurrency=10 and handlers that park on a
-Promise. OpenClaw side spawns 10 parallel `openclaw agent --local`
-processes and sums their RSS via `ps -o rss=`.
-
-Handlers are intentionally cheap sleeps — we measure harness memory,
-not LLM client state. The LLM client state would be comparable on both
-sides.
-
-**Minions costs 2 MB to keep 10 subagents in flight. OpenClaw costs
-814 MB. At scale, this difference decides whether you can run 10
-subagents or 100 on the same machine.**
-
-Source: `test/e2e/bench-vs-openclaw/memory.bench.ts`
-
-## What this means for a Minions user
-
-If you have a script today that spawns `openclaw agent --local` N times,
-every one of these numbers gets better when you move to Minions:
-
-- **Crash and your work doesn't vanish.** Worker dies, PG keeps the
-  row, another worker picks it up. Zero extra code on your side.
-- **Per-dispatch wall time drops ~10×** because the worker stays warm.
-  Process startup is where your time was going, not the LLM.
-- **Fan-out scales past 10-wide without you hand-tuning concurrency.**
-  Worker does the throttling; the queue does the durability. OC
-  parallel spawn hits a 40% failure wall around 10-wide on this hardware.
-- **Memory stops being the bottleneck.** 2 MB per in-flight job vs
-  ~80 MB per process changes what "10 concurrent subagents" costs you
-  on a box.
-
-## What this doesn't say
-
-- We didn't test OpenClaw's gateway multi-agent mode. If you run the
-  gateway, you get persistent agent state across turns, real multi-agent
-  routing, and different cost characteristics. The gateway is OC's
-  production mode, and we're not claiming Minions beats it at what it
-  does. We're saying: if your pattern is "dispatch a subagent, get a
-  reply, maybe do this 10 times," the `--local` CLI is what you're
-  reaching for, and Minions beats it by ~10-400× depending on the axis.
-- We didn't run under load (100s of concurrent jobs, hours of sustained
-  work). These are observational point measurements, not a stress test.
-- We ran claude-haiku-4-5. For slower/larger models the absolute
-  numbers shift but the ratios stay roughly the same — the overhead
-  is process boot and persistence, not model size.
-
-## Reproducing
-
-```bash
-# 1. Start a test Postgres
-docker run -d --name gbrain-test-pg \
-  -e POSTGRES_USER=postgres -e POSTGRES_PASSWORD=postgres \
-  -e POSTGRES_DB=gbrain_test \
-  -p 5436:5432 pgvector/pgvector:pg16
-
-# 2. Set env
-export DATABASE_URL=postgresql://postgres:postgres@localhost:5436/gbrain_test
-export ANTHROPIC_API_KEY=sk-ant-...
-
-# 3. Run each bench (durability + memory are free; throughput + fan-out
-#    cost ~$0.25 in claude-haiku-4-5 tokens total)
-bun test ./test/e2e/bench-vs-openclaw/durability.bench.ts
-bun test ./test/e2e/bench-vs-openclaw/throughput.bench.ts
-bun test ./test/e2e/bench-vs-openclaw/fanout.bench.ts
-bun test ./test/e2e/bench-vs-openclaw/memory.bench.ts
-
-# 4. Tear down
-docker stop gbrain-test-pg && docker rm gbrain-test-pg
-```
-
-## One-line summary
-
-Minions rescues 10/10 jobs from a crash in under half a second while
-OpenClaw `--local` loses all of them; it delivers each dispatch ~10×
-faster, fans out 10-wide in ~1 second vs ~22 seconds at 43% OC failure
-rate, and holds 10 in-flight subagents in 2 MB vs 814 MB.
diff --git a/docs/benchmarks/2026-04-18-tweet-ingestion.md b/docs/benchmarks/2026-04-18-tweet-ingestion.md
deleted file mode 100644
index 497f8afa6..000000000
--- a/docs/benchmarks/2026-04-18-tweet-ingestion.md
+++ /dev/null
@@ -1,176 +0,0 @@
-# Tweet Ingestion Benchmark: Minions vs OpenClaw Sub-agents
-
-**Date:** 2026-04-18
-**Branch:** garrytan/minions-jobs
-**Suite:** `test/e2e/bench-vs-openclaw/tweet-ingest.bench.ts`
-**Minions:** v0.11.0 (PR #130)
-**OpenClaw:** 2026.4.10
-**Model:** none (Minions) vs anthropic/claude-sonnet-4 (OpenClaw)
-
-## Why this benchmark exists
-
-The existing throughput/fanout/durability benchmarks use a trivial LLM
-prompt ("Reply with just: OK"). They measure queue overhead, not real work.
-
-This benchmark measures a **real production task**: pull a month of tweets
-from the X API, parse them into a structured brain page, git commit, and
-sync to gbrain. This is work that an agent does every day. It's
-deterministic — same input always produces the same steps in the same
-order. The question: should deterministic brain-write work go through an
-LLM (sub-agent) or through code (Minions)?
-
-## Methodology
-
-**Task:** Pull ~100 my social posts for one month from the X full-archive
-search API, write a markdown brain page with frontmatter + engagement
-metrics + tweet links, git commit, and submit a `gbrain sync` job.
-
-**Minions side:** A TypeScript function that:
-1. `fetch()` the X API (one HTTP call)
-2. `JSON.parse()` → `writeFileSync()` the brain page
-3. `execSync('git commit')`
-4. `queue.add('sync', { repo, noPull: true })`
-
-No LLM involved. The handler is code. Total overhead on top of I/O:
-queue add + git commit.
-
-**OpenClaw side:** Spawn `openclaw agent --local` with a task prompt that
-describes the same pipeline in English. The model (claude-sonnet-4):
-1. Reads the task, plans approach
-2. Calls `exec` tool for curl
-3. Calls `exec` tool for python (parse + write)
-4. Calls `exec` tool for git commit
-5. Reports result
-
-Same work, but the model decides each step.
-
-**Runs:** 5 serial per method. Each run uses a different month (2020-07
-through 2020-11) to avoid caching effects. Pages are cleaned up after.
-
-**Environment:** Tested on a production Render container (ephemeral, ARM64)
-with Supabase Postgres (us-east-1) and a 45K-page brain. Also
-reproducible on localhost with Docker Postgres — see instructions below.
-
-## Honest caveats
-
-- **X API latency varies.** The X full-archive search endpoint takes
-  200-500ms depending on load. Both sides pay this equally. We're
-  measuring the PIPELINE overhead, not the API.
-- **OpenClaw `--local` is not the gateway.** The gateway has persistent
-  sessions, tool caching, and context reuse. `--local` is the scripted
-  dispatch path — what you'd use in a cron job or automation script.
-  That's the apples-to-apples comparison for deterministic work.
-- **The sub-agent has to figure out the same pipeline every time.**
-  That's the core inefficiency: spending tokens for the model to
-  rediscover steps that never change. With Minions, the steps are code.
-- **N=5 is small.** Enough to see the order-of-magnitude delta, not
-  enough to prove tight tails. Run N=20 for statistical significance.
-
-## Results
-
-### Minions (5 runs, serial)
-
-| Run | Month | Tweets | Wall time | Status |
-|-----|-------|--------|-----------|--------|
-| 1 | 2020-07 | 99 | 753ms | ✅ |
-| 2 | 2020-08 | 87 | 681ms | ✅ |
-| 3 | 2020-09 | 92 | 724ms | ✅ |
-| 4 | 2020-10 | 78 | 698ms | ✅ |
-| 5 | 2020-11 | 103 | 741ms | ✅ |
-
-**Stats:** mean=719ms p50=724ms p95=753ms min=681ms max=753ms
-**Success rate:** 5/5 (100%)
-**Token cost:** $0.00
-
-### OpenClaw Sub-agent (5 runs, serial)
-
-| Run | Month | Tweets | Wall time | Status |
-|-----|-------|--------|-----------|--------|
-| 1 | 2020-07 | — | >10,000ms | ❌ gateway timeout |
-| 2 | 2020-08 | — | >10,000ms | ❌ gateway timeout |
-| 3 | 2020-09 | 99 | 12,340ms | ✅ |
-| 4 | 2020-10 | 87 | 11,890ms | ✅ |
-| 5 | 2020-11 | 92 | 13,210ms | ✅ |
-
-**Stats (successful only):** mean=12,480ms p50=12,340ms
-**Success rate:** 3/5 (60%) — 2 gateway timeouts under production load
-**Token cost:** ~$0.03 per successful run × 3 = $0.09
-
-> **Note:** Gateway timeouts occurred because the production OpenClaw
-> instance was running 19 active cron jobs + heartbeats. The gateway's
-> session spawn queue was saturated. This is a realistic production
-> scenario, not an artificial constraint.
-
-### Comparison
-
-| Metric | Minions | OpenClaw Sub-agent | Ratio |
-|--------|---------|-------------------|-------|
-| **Mean wall time** | **719ms** | **12,480ms** | **17.3×** |
-| **p50** | 724ms | 12,340ms | 17.0× |
-| **Success rate** | 100% | 60% | — |
-| **Token cost per run** | $0.00 | ~$0.03 | ∞ |
-| **Survives restart** | ✅ | ❌ | — |
-| **Progress tracking** | ✅ `jobs get` | ❌ | — |
-| **Auto-retry** | ✅ 3 attempts | ❌ | — |
-
-### At scale: 36-month backfill
-
-We also measured a real backfill: pull 36 months of tweets (2021-2023,
-19,240 tweets total) and ingest each month as a brain page.
-
-| Metric | Minions | OpenClaw Sub-agent (est.) |
-|--------|---------|--------------------------|
-| **Total time** | ~15 min | ~7.5 min (best case) to ∞ (gateway timeouts) |
-| **Total cost** | $0.00 | ~$1.08 (36 × $0.03) |
-| **Expected failures** | 0 | ~14 (36 × 40% failure rate) |
-| **Manual intervention** | None | Re-spawn failed months |
-
-The Minions path completed all 36 months unattended. The sub-agent path
-would require monitoring and re-spawning failures.
-
-## The routing insight
-
-This benchmark measures **deterministic work** — work where the steps
-never change regardless of input. Pull → parse → write → commit → sync.
-The same pipeline every time. Spending $0.03 and 12 seconds for a model
-to rediscover these steps is waste.
-
-The routing rule that falls out of this data:
-
-> **Deterministic** (same input → same steps → same output) → **Minions**
-> Zero tokens. Sub-second. Durable. Auto-retry.
->
-> **Judgment** (input requires assessment/decision) → **Sub-agents**
-> Model decides what to do. Worth the token cost.
-
-Examples:
-- Tweet ingestion → Minions (always the same pipeline)
-- Calendar sync → Minions (always the same pipeline)
-- Email triage → Sub-agent (model decides priority + reply)
-- Meeting prep → Sub-agent (model synthesizes briefing)
-
-## Reproducing
-
-```bash
-# 1. Set environment
-export X_BEARER_TOKEN=...           # external API bearer token
-export DATABASE_URL=postgresql://... # Postgres with gbrain schema v7+
-export BRAIN_PATH=/path/to/brain    # Git repo with brain pages
-export ANTHROPIC_API_KEY=sk-ant-... # For OpenClaw side only
-
-# 2. Run the benchmark
-bun test test/e2e/bench-vs-openclaw/tweet-ingest.bench.ts
-
-# 3. Cost: ~$0.15 total (5 OC runs × ~$0.03 each, Minions = $0)
-
-# 4. On localhost without X API: mock the fetch in the test file
-#    to return a canned JSON response. The benchmark measures
-#    pipeline overhead, not API latency.
-```
-
-## One-line summary
-
-Minions ingests a month of tweets in 719ms for $0 with 100% reliability.
-OpenClaw sub-agents take 12.5 seconds, cost $0.03, and fail 40% of the
-time under production load. For deterministic brain-write work, Minions
-is 17× faster, infinitely cheaper, and categorically more reliable.
diff --git a/docs/benchmarks/2026-04-19-knowledge-runtime-v0.13.md b/docs/benchmarks/2026-04-19-knowledge-runtime-v0.13.md
deleted file mode 100644
index 9eae77202..000000000
--- a/docs/benchmarks/2026-04-19-knowledge-runtime-v0.13.md
+++ /dev/null
@@ -1,190 +0,0 @@
-# Knowledge Runtime v0.13 — Benchmark Deltas
-
-What this branch actually changes, measured. All numbers are reproducible from
-the scripts in `test/`. No real-world traffic, no API keys, no private data.
-
-**Headline:** Step B (auto-timeline on put_page) is the only change that moves
-benchmark numbers, and it moves them from 0% to 100% on the one metric that
-matters for agent workflow: "can I query the timeline right after I wrote the
-page?"
-
-The retrieval-quality benchmarks (graph-quality, search-quality) are unchanged
-because this branch didn't touch the search or graph-query hot paths. That's
-the expected result and it's the proof that the knowledge-runtime work didn't
-regress anything it wasn't supposed to change.
-
----
-
-## Benchmark 1: put_page latency
-
-**Script:** `bun run test/benchmark-put-page-latency.ts --json`
-**Load:** 200 `put_page` operation calls against PGLite in-process, half
-carrying 3 timeline entries, 10 seed target pages for auto-link to resolve.
-
-|  | master (v0.12.1, c0b6219) | branch (v0.13.0.0) | Δ |
-|---|---:|---:|---:|
-| mean | 2.00 ms | 2.58 ms | **+0.58 ms (+29%)** |
-| p50 | 1.92 ms | 2.31 ms | +0.39 ms (+20%) |
-| p95 | 2.56 ms | 3.57 ms | +1.01 ms (+39%) |
-| p99 | 3.46 ms | 13.44 ms | +9.98 ms (+288%) |
-| max | 10.89 ms | 14.34 ms | +3.45 ms |
-| timeline entries extracted | **0** | **300** | +300 |
-
-**Read:** Step B adds ~0.5 ms to mean `put_page` latency and the branch now
-extracts 300 timeline entries across 200 writes for free. Master does zero.
-The absolute cost is invisible in any practical workflow. The p99 tail
-doubled (3.5 → 13.4 ms); absolute is still <15 ms and almost certainly
-batch-flush variance, not a regression worth acting on.
-
----
-
-## Benchmark 2: Time-to-queryable brain
-
-**Script:** `bun run test/benchmark-knowledge-runtime.ts --json` (section `ttq`)
-**Scenario:** 20 pages ingested via the `put_page` OPERATION (not the engine
-method). 40 expected timeline entries across them. Immediately after ingest,
-query `engine.getTimeline(slug)` for each expected entry.
-
-|  | queryable right after ingest |
-|---|---:|
-| branch (auto_timeline on, default) | **40/40 (100%)** |
-| master (auto_timeline off, current behavior) | 0/40 (0%) |
-
-**Read:** On master, zero timeline queries return answers after a write. The
-user has to remember to run `gbrain extract timeline` as a second step or
-their agent gets blank results. On branch, every timeline query works the
-moment the page lands. This is the "boil-the-lake" principle in action: when
-AI makes the marginal cost near-zero, always do the complete thing.
-
----
-
-## Benchmark 3: Integrity repair rate (mocked resolver)
-
-**Script:** `bun run test/benchmark-knowledge-runtime.ts --json` (section `integrity`)
-**Scenario:** 50 pages seeded with bare-tweet phrases and `x_handle`
-frontmatter. Fake `x_handle_to_tweet` resolver returns confidence deterministically
-from a 70/20/10 distribution (70% high, 20% mid, 10% low). Three-bucket
-repair logic runs the same way `gbrain integrity auto` does in production.
-
-|  | count | % |
-|---|---:|---:|
-| auto-repair (confidence ≥ 0.8) | 35 | 70% |
-| review queue (0.5 ≤ c < 0.8) | 10 | 20% |
-| skip (c < 0.5) | 5 | 10% |
-
-**Read:** Master has no integrity repair at all — this feature is new in
-v0.13. The machinery delivers exactly the three-bucket split the design
-promised. With the real X API the absolute numbers will shift depending on
-how well the resolver discriminates, but the pipeline is provably correct.
-Zero phrases slip through without a confidence-bucketed decision.
-
----
-
-## Benchmark 4: Doctor signal completeness
-
-**Script:** `bun run test/benchmark-knowledge-runtime.ts --json` (section `doctor`)
-**Scenario:** Seed a brain with 7 known issues: 3 bare-tweet phrases across
-2 pages (one-hit-per-line rule reduces this to 2 surfaceable), 3 external
-link citations, 1 grandfathered page (frontmatter `validate: false`, which
-should be skipped). Run the `scanIntegrity` helper that doctor now invokes
-in non-fast mode.
-
-|  | count |
-|---|---:|
-| issues planted | 7 |
-| should surface | 6 |
-| grandfathered (correctly skipped) | 1 |
-| **surfaced** | **5 (83%)** |
-| bare tweets caught | 2/2 lines |
-| external links caught | 3/3 |
-| grandfathered page respected | 1/1 |
-
-**Read:** Master's `gbrain doctor` catches zero of these — doctor had no
-integrity awareness before this branch. Now it surfaces 100% of the
-surfaceable issues and correctly respects the grandfather flag. The 83%
-headline comes from the planted-vs-surfaceable counting: 7 planted, 1 opted
-out, 6 should surface, 5 did. In terms of detection rate for real issues,
-it's 5/5 on lines that have bare-tweet content.
-
----
-
-## Benchmarks that did NOT move (proof of no regression)
-
-### Graph quality benchmark
-
-**Script:** `bun run test/benchmark-graph-quality.ts --json`
-**Load:** 80 fictional pages, 35 relational queries across 7 categories.
-
-| metric | master | branch | Δ |
-|---|---:|---:|---|
-| link_recall | 0.889 | 0.889 | 0 |
-| link_precision | 1.000 | 1.000 | 0 |
-| type_accuracy | 0.889 | 0.889 | 0 |
-| timeline_recall | 1.000 | 1.000 | 0 |
-| timeline_precision | 1.000 | 1.000 | 0 |
-| relational_recall | 0.900 | 0.900 | 0 |
-| relational_precision | 1.000 | 1.000 | 0 |
-| idempotent_links | true | true | = |
-| idempotent_timeline | true | true | = |
-
-**Read:** Identical. The benchmark uses `engine.putPage()` + explicit
-`runExtract` calls, which bypass the operation handler where Step B lives.
-That's why the numbers don't move, and that's the right outcome: the graph
-layer's extraction quality hasn't changed, only the ingest ergonomics.
-
-### Search quality benchmark
-
-**Script:** `bun run test/benchmark-search-quality.ts`
-**Load:** 30 pages, 20 queries with graded relevance. Modes A (baseline),
-B (boost only), C (boost + intent classifier).
-
-| metric | A (baseline) | B (boost) | C (full) | Δ master→branch |
-|---|---:|---:|---:|---|
-| P@1 | 0.947 | 0.895 | 0.947 | 0 |
-| P@5 | 0.811 | 0.674 | 0.695 | 0 |
-| MRR | 0.974 | 0.939 | 0.974 | 0 |
-| nDCG@5 | 1.191 | 1.028 | 1.069 | 0 |
-
-**Read:** Identical across all three modes. Search scoring is decided by
-hybrid search + RRF + dedup, none of which this branch touched.
-
----
-
-## Reproducing these numbers
-
-```bash
-# From this branch
-bun run test/benchmark-put-page-latency.ts --json
-bun run test/benchmark-knowledge-runtime.ts --json
-bun run test/benchmark-graph-quality.ts --json
-bun run test/benchmark-search-quality.ts
-
-# Compare against master
-cd /path/to/gbrain-master-worktree
-# (copy benchmark-put-page-latency.ts and benchmark-knowledge-runtime.ts
-# over if they're not on master yet; they're the new scripts)
-bun run test/benchmark-put-page-latency.ts --json
-bun run test/benchmark-graph-quality.ts --json
-bun run test/benchmark-search-quality.ts
-```
-
-All four scripts run in-process against PGLite. No network, no external DB,
-no API keys. They complete in under 30 seconds combined.
-
----
-
-## Bottom line
-
-| benchmark | moves? | direction |
-|---|---|---|
-| put_page latency | yes | +0.5ms cost for 300 free timeline entries per 200 writes |
-| time-to-queryable | yes | 0% → 100% |
-| integrity repair rate | new | n/a on master, 70/20/10 split delivered |
-| doctor completeness | new | 0% → 100% on real issues |
-| graph quality | no | unchanged, as designed |
-| search quality | no | unchanged, as designed |
-
-The branch does what it said it would do. The retrieval benchmarks stay flat
-and the ingest/repair/health benchmarks move from zero to working. That's
-the shape of a good platform change: one new dimension opens up, existing
-dimensions don't regress.
diff --git a/eval/data/world-v1/_ledger.json b/eval/data/world-v1/_ledger.json
deleted file mode 100644
index 0a5fd8cef..000000000
--- a/eval/data/world-v1/_ledger.json
+++ /dev/null
@@ -1,13 +0,0 @@
-{
-  "generated_at": "2026-04-18T04:13:16.027Z",
-  "model": "claude-opus-4-5",
-  "pricing": {
-    "input_per_m": 15,
-    "output_per_m": 75
-  },
-  "inputTokens": 18359,
-  "outputTokens": 38228,
-  "costUsd": 3.1424849999999998,
-  "calls": 49,
-  "files_total": 240
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__accel-5.json b/eval/data/world-v1/companies__accel-5.json
deleted file mode 100644
index cb8df4d59..000000000
--- a/eval/data/world-v1/companies__accel-5.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "companies/accel-5",
-  "type": "company",
-  "title": "Accel - Global Venture Capital Firm",
-  "compiled_truth": "Accel is one of the most established venture capital firms in the world, with a track record spanning over four decades. Founded in 1983, the firm has evolved from a Silicon Valley stalwart into a truly global operation with offices in Palo Alto, London, and Bangalore. They've backed some of the most consequential technology companies of the past two decades, including Facebook, Spotify, Slack, and Dropbox.\n\nThe firm operates across multiple stages, though they're perhaps best known for their Series A and Series B investments. Accel manages billions in assets across various funds, with recent vintages exceeding $3 billion for their US and Europe-focused vehicles. Their investment thesis tends to favor founders building category-defining companies in enterprise software, consumer tech, fintech, and increasingly, AI infrastructure.\n\nAccel's partnership model emphasizes deep sector expertise. Partners like Sonali De Rycker have built formidable reputations in European fintech, while others focus on developer tools or consumer applications. The firm has been notably active in the generative AI wave, making early bets on companies building foundational models and application layers. They've developed strong relationships with accelerators like [Y Combinator](companies/y-combinator) and often co-invest alongside firms such as [Andreessen Horowitz](companies/a16z) on competitive deals.\n\nRecent years have seen Accel double down on international expansion. Their India fund has become one of the most active institutional investors in the subcontinent, backing companies like Flipkart and Swiggy before they became household names. The London office continues to punch above its weight in European tech circles.\n\nThe firm's culture is often described as founder-friendly but rigorous. They're known for taking board seats seriously and providing operational support beyond just capital. Accel's brand carries significant weight in fundraising conversations—a term sheet from them often signals quality to follow-on investors. Critics sometimes note their portfolio can feel conservative compared to newer entrants, but longevity has its advantages. They've seen multiple market cycles and tend to maintain disciplined valuations even in frothy markets.",
-  "timeline": [
-    "- **2021-03-15** | Accel closes $3 billion early-stage fund, largest in firm history at the time",
-    "- **2021-09-22** | Led Series B for enterprise AI startup alongside [Andreessen Horowitz](companies/a16z)",
-    "- **2022-04-10** | Opens expanded London office to support growing European portfolio",
-    "- **2022-11-08** | Partner Rich Wong speaks at Web Summit on enterprise software trends",
-    "- **2023-02-14** | Announces $650 million India-focused fund, sixth in the region",
-    "- **2023-08-30** | Leads seed round for [Y Combinator](companies/y-combinator) batch company building AI code review tools",
-    "- **2024-01-19** | Accel publishes annual Euroscape report showing record European unicorn creation",
-    "- **2024-06-05** | Makes significant investment in robotics startup focused on warehouse automation",
-    "- **2025-02-11** | Closes latest growth fund at $4.2 billion amid competitive fundraising environment",
-    "- **2025-09-03** | Hosts annual CEO summit in Portofino, bringing together 80+ portfolio founders"
-  ],
-  "_facts": {
-    "type": "company",
-    "slug": "companies/accel-5",
-    "name": "Accel",
-    "category": "vc",
-    "industry": "venture capital"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__acme-0.json b/eval/data/world-v1/companies__acme-0.json
deleted file mode 100644
index ee98179f5..000000000
--- a/eval/data/world-v1/companies__acme-0.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "companies/acme-0",
-  "type": "company",
-  "title": "Acme",
-  "compiled_truth": "Acme is a robotics startup founded in 2021 by [Mia Brown](people/mia-brown-0), who previously spent nearly a decade in industrial automation before striking out on her own. The company focuses on developing modular robotic systems for small and mid-sized warehouses—an underserved market segment that larger players have largely ignored. Their flagship product, the Acme Flex Unit, is a mobile picking robot that can be deployed in facilities without major infrastructure changes.\n\nThe startup has attracted notable backing from angel investors including [Chris Jackson](people/chris-jackson-91) and [Ian Anderson](people/ian-anderson-105), both of whom participated in the seed round closed in early 2022. Jackson in particular has been hands-on, joining several board meetings and making introductions to potential enterprise customers. Acme raised a modest $2.3M initially, deliberately staying lean while proving out the core technology.\n\nMia Brown serves as CEO and remains deeply involved in product development. She's known for an engineering-first approach to company building, often spending time on the factory floor alongside her small team. The company currently employs around 25 people, mostly engineers, operating out of a converted warehouse space in Austin. Acme has been quiet about expansion plans but insiders suggest a Series A is in the works for late 2025.\n\nThe robotics market is crowded, yet Acme has carved out a niche by targeting businesses too small for enterprise solutions but too large for manual operations alone. Early customers include regional e-commerce fulfillment centers and a few specialty food distributors. Retention has been strong, with several pilots converting to full deployments.\n\nRecent moves include a partnership with a logistics software provider to integrate Acme's robots into broader warehouse managment systems. The company also hired its first dedicated sales lead in Q1 2025, signaling a shift toward scaling comercial operations. Despite limited public visibility, Acme has built a reputation in robotics circles for reliable hardware and responsive support.",
-  "timeline": "- **2021-06-15** | Acme incorporated in Delaware by [Mia Brown](people/mia-brown-0)\n- **2022-02-10** | Closed $2.3M seed round led by [Chris Jackson](people/chris-jackson-91) and [Ian Anderson](people/ian-anderson-105)\n- **2022-09-01** | First prototype of Acme Flex Unit completed\n- **2023-03-22** | Signed pilot agreement with regional fulfillment center in Texas\n- **2023-11-08** | Expanded team to 15 employees, opened Austin facility\n- **2024-04-17** | Converted three pilot customers to full commercial deployments\n- **2024-10-30** | Announced integration partnership with WarehouseOS software platform\n- **2025-01-14** | Hired first dedicated head of sales, marking commercial scale-up\n- **2025-06-02** | [Mia Brown](people/mia-brown-0) spoke at RoboTech Summit on modular automation\n- **2025-11-20** | Series A discussions reportedly underway with multiple VC firms",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/acme-0",
-    "name": "Acme",
-    "category": "startup",
-    "industry": "robotics",
-    "founded_year": 2021,
-    "founders": [
-      "people/mia-brown-0"
-    ],
-    "investors": [
-      "people/chris-jackson-91",
-      "people/ian-anderson-105"
-    ],
-    "employees": [
-      "people/chris-smith-110"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__acme-labs-50.json b/eval/data/world-v1/companies__acme-labs-50.json
deleted file mode 100644
index f85187890..000000000
--- a/eval/data/world-v1/companies__acme-labs-50.json
+++ /dev/null
@@ -1,27 +0,0 @@
-{
-  "slug": "companies/acme-labs-50",
-  "type": "company",
-  "title": "Acme Labs",
-  "compiled_truth": "Acme Labs is a cybersecurity startup founded in 2019 by [Ian Kim](people/ian-kim-50), a serial entrepreneur with deep roots in enterprise security software. The company emerged from Kim's frustration with legacy endpoint protection tools that couldn't keep pace with modern threat vectors. Based out of Austin, Texas, Acme has grown from a three-person operation to a team of roughly 45 engineers and security researchers.\n\nThe company's flagship product is a real-time threat detection platform that uses behavioral analysis to identify anomalies before they escalate into full breaches. Unlike traditional signature-based approaches, Acme's system learns the normal patterns of network traffic and user behavior, flagging deviations that might indicate compromise. Early customers were mid-market financial services firms, though the company has since expanded into healthcare and logistics verticals.\n\nFunding came relatively early. [Helen Martinez](people/helen-martinez-87) led the seed round in late 2020, bringing not just capital but also her extensive network in enterprise software distribution. Martinez has remained closely involved, attending board meetings and occasionally making introductions to potential strategic partners. The Series A followed in 2022, though terms were not publicly disclosed.\n\nOn the advisory side, [Wendy Wilson](people/wendy-wilson-170) joined in 2021 to help shape go-to-market strategy. Wilson's backgorund in scaling B2B SaaS companies proved invaluable as Acme transitioned from founder-led sales to a more structured revenue organization. She's credited with pushing the team to focus on a narrower ICP rather than chasing every inbound lead.\n\nAcme Labs has built a reputation for technical depth. Their engineering blog regularly publishes threat research, and several team members speak at conferences like DEF CON and BSides. The culture leans scrappy—Kim is known for keeping overhead low and reinvesting heavily into R&D. Recent chatter suggests the company is exploring an AI-powered SOC assistant, though nothing has been formally anounced. Competition remains fierce from both established players and well-funded startups, but Acme's focus on mid-market customers gives them a defensible niche.",
-  "timeline": "- **2019-03-12** | Acme Labs incorporated in Delaware; [Ian Kim](people/ian-kim-50) begins building initial prototype\n- **2019-11-04** | First paying customer signed — a regional credit union in Texas\n- **2020-09-18** | Seed round closed with [Helen Martinez](people/helen-martinez-87) leading the investment\n- **2021-02-22** | [Wendy Wilson](people/wendy-wilson-170) joins as strategic advisor\n- **2021-08-30** | Acme releases v2.0 of threat detection platform with behavioral analytics engine\n- **2022-04-15** | Series A funding completed; team expands to 30 employees\n- **2023-06-09** | Ian Kim delivers keynote at RSA Conference on zero-trust architecture\n- **2024-01-17** | Partnership announced with major SIEM vendor for native integration\n- **2024-11-03** | Acme Labs crosses $10M ARR milestone\n- **2025-07-21** | Internal demo of AI-powered SOC assistant shown to select customers",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/acme-labs-50",
-    "name": "Acme Labs",
-    "category": "startup",
-    "industry": "cybersecurity",
-    "founded_year": 2019,
-    "founders": [
-      "people/ian-kim-50"
-    ],
-    "investors": [
-      "people/helen-martinez-87"
-    ],
-    "employees": [
-      "people/vera-martinez-160"
-    ],
-    "advisors": [
-      "people/wendy-wilson-170"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__amazon-3.json b/eval/data/world-v1/companies__amazon-3.json
deleted file mode 100644
index 2a5101afb..000000000
--- a/eval/data/world-v1/companies__amazon-3.json
+++ /dev/null
@@ -1,15 +0,0 @@
-{
-  "slug": "companies/amazon-3",
-  "type": "company",
-  "title": "Amazon - Cybersecurity Acquirer",
-  "compiled_truth": "Amazon, founded in 1998, has evolved far beyond its origins as an online bookstore to become one of the most formidable players in the technology sector. While most know the company for its e-commerce dominance and AWS cloud infrastructure, Amazon has quietly built a substantial presence in cybersecurity through strategic acquisitions and internal development.\n\nThe company's approach to cybersecurity M&A has been methodical and often under the radar. Rather than making splashy billion-dollar deals that attract media attention, Amazon tends to acquire smaller, specialized firms that can be integrated into its existing AWS security stack. This strategy allows them to enhance offerings like AWS Shield, GuardDuty, and Security Hub without the integration headaches that plague larger mergers.\n\nAmazon's cybersecurity ambitions are driven partly by necesity—protecting its massive cloud infrastructure and the millions of businesses that depend on it requires constant innovation. The company processes an astronomical volume of security events daily, giving it unique datasets for training threat detection models. Some industry observers beleive this data advantage makes Amazon a sleeping giant in the security space.\n\nRecent moves suggest the company is getting more aggressive. They've been spotted at major security conferences with larger acquisition teams, and rumors persist about interest in several endpoint detection startups. The hiring of former NSA and CISA officials into senior AWS security roles signals a maturation of their strategy.\n\nCompetition with [Microsoft](companies/microsoft) in the cloud security space has intensified, with both giants racing to offer comprehensive security platforms that reduce customers' need for third-party tools. Amazon's relationship with specialized security vendors is complicated—they partner with many through the AWS Marketplace while simultaneously building competing capabilities.\n\nThe firm maintains close ties with government contractors and has pursued FedRAMP certifications aggressively. Their work with [Palantir](companies/palantir) on certain government cloud initiatives demonstrates Amazon's willingness to collaborate when strategic interests align, though the relationship has had its tense moments over competing contract bids.",
-  "timeline": "- **2021-03-15** | Amazon acquires small threat intelligence startup for undisclosed sum, team absorbed into AWS Security division\n- **2021-09-22** | Launched AWS Security Lake at re:Invent, consolidating security data management capabilities\n- **2022-04-08** | Hired former CISA deputy director to lead government security initiatives\n- **2022-11-30** | Announced expanded partnership with [Microsoft](companies/microsoft) on cross-cloud security standards, surprising industry observers\n- **2023-06-14** | Acquisition of Israeli-based API security firm closes, adding to AppSec portfolio\n- **2023-12-01** | AWS Security Hub surpasses 50,000 enterprise customers milestone\n- **2024-05-19** | Internal memo leaked showing renewed focus on endpoint security acquisitions\n- **2024-10-03** | Joint threat intelligence sharing agreement signed with [Palantir](companies/palantir) for federal contracts\n- **2025-02-28** | Rumored in late-stage talks with two identity management startups\n- **2025-08-11** | Opened dedicated cybersecurity R&D center in Austin, Texas",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/amazon-3",
-    "name": "Amazon",
-    "category": "acquirer",
-    "industry": "cybersecurity",
-    "founded_year": 1998
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__anchor-28.json b/eval/data/world-v1/companies__anchor-28.json
deleted file mode 100644
index d7dfc9764..000000000
--- a/eval/data/world-v1/companies__anchor-28.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "companies/anchor-28",
-  "type": "company",
-  "title": "Anchor - Data Infrastructure Startup",
-  "compiled_truth": "Anchor is a data infrastructure startup founded in 2021 by [Carol Wilson](people/carol-wilson-28), a veteran engineer who previously spent nearly a decade building distributed systems at major tech companies. The company focuses on solving one of the most persistent problems in modern data stacks: reliable data synchronization across heterogenous cloud environments.\n\nThe core product is a managed service that handles bi-directional sync between data warehouses, operational databases, and third-party SaaS tools. Unlike traditional ETL pipelines, Anchor's approach treats data synchronization as a continous process rather than batch jobs, enabling near real-time consistency across systems. This has proven particularly valuable for companies running hybrid cloud architectures or those mid-migration between legacy systems and modern infrastructure.\n\nAnchor raised its seed round from [Sarah Williams](people/sarah-williams-92) and [Kate Anderson](people/kate-anderson-107), both of whom have deep backgrounds in enterprise software investing. The round closed in early 2022 and allowed the company to expand beyond its initial three-person team. Sarah Williams in particular has been an active board observer, reportedly helping Anchor navigate early enterprise sales conversations.\n\nThe startup has been deliberatly quiet about customer names, though industry observers have noted several mid-market fintech companies using Anchor's sync layer for compliance-related data requirements. Carol Wilson has spoken at a handful of data engineering conferences about the technical challenges of conflict resolution in distributed data systems—talks that have helped establish Anchor's credibility in a crowded market.\n\nGrowth has been steady if not explosive. The company operates with a lean team, currently around fifteen employees, mostly engineers. There's been some speculation about a Series A in 2024, though nothing confirmed publically. Anchor competes with larger players like Fivetran and Airbyte, but differentiates on the bi-directional sync capabilities and lower latency guarantees. The data infrastructure space remains intensely competitive, but Anchor has carved out a defensible niche.",
-  "timeline": "- **2021-03-15** | Anchor incorporated in Delaware by [Carol Wilson](people/carol-wilson-28)\n- **2021-06-22** | First working prototype of bi-directional sync engine completed\n- **2022-01-18** | Closed seed round led by [Sarah Williams](people/sarah-williams-92) and [Kate Anderson](people/kate-anderson-107)\n- **2022-08-03** | Launched private beta with five design partners\n- **2023-02-11** | Carol Wilson delivered keynote on distributed sync at DataEngConf Austin\n- **2023-07-29** | General availability launch; pricing tiers announced\n- **2023-11-14** | Reached 50 paying customers milestone\n- **2024-04-08** | Opened second office in Denver for engineering expansion\n- **2024-09-22** | Partnership announced with major cloud provider (details under NDA)\n- **2025-01-30** | Anchor featured in industry report on emerging data infrastructure vendors",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/anchor-28",
-    "name": "Anchor",
-    "category": "startup",
-    "industry": "data infrastructure",
-    "founded_year": 2021,
-    "founders": [
-      "people/carol-wilson-28"
-    ],
-    "investors": [
-      "people/sarah-williams-92",
-      "people/kate-anderson-107"
-    ],
-    "employees": [
-      "people/tara-hernandez-138"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__andreessen-horowitz-2.json b/eval/data/world-v1/companies__andreessen-horowitz-2.json
deleted file mode 100644
index 2d8c79198..000000000
--- a/eval/data/world-v1/companies__andreessen-horowitz-2.json
+++ /dev/null
@@ -1,14 +0,0 @@
-{
-  "slug": "companies/andreessen-horowitz-2",
-  "type": "company",
-  "title": "Andreessen Horowitz",
-  "compiled_truth": "Andreessen Horowitz, widely known as a16z, is one of the most influential venture capital firms in Silicon Valley and arguably the world. Founded in 2009 by Marc Andreessen and Ben Horowitz, the firm has grown from a scrappy upstart challenging the old guard of VC into a multi-billion dollar asset manager with funds spanning crypto, bio, games, and traditional enterprise software.\n\nThe firm's thesis has always been rooted in the belief that software is eating the world—a phrase Marc coined in his famous 2011 Wall Street Journal essay. This conviction drove early bets on companies like Facebook, Twitter, Airbnb, and Coinbase, generating massive returns for limited partners. a16z pioneered the \"founder-friendly\" approach to venture capital, offering not just capital but an entire platform of services: recruiting, marketing, executive coaching, and regulatory expertise.\n\nIn recent years, Andreessen Horowitz has leaned heavily into crypto and web3, raising multiple dedicated funds totaling billions of dollars. This bet has been controversial—critics argue the firm is too bullish on speculative assets, while supporters see it as visionary positioning for the next computing platform. The firm also expanded into consumer health through a16z Bio and doubled down on American Dynamism, a thesis around backing companies building in defense, aerospace, and manufacturing.\n\nThe partnership includes heavyweights like Chris Dixon (leading crypto), Vijay Pande (bio), and Andrew Chen (consumer). Marc remains a polarizing figure on social media, often wading into political and cultural debates that generate significant attention. Some view this as distraction, others as authentic engagement. Ben Horowitz has focused more on cultural content, including his popular book \"The Hard Thing About Hard Things.\"\n\na16z competes fiercely with firms like [Sequoia Capital](companies/sequoia-capital) and [General Catalyst](companies/general-catalyst) for the best deals. Their approach to content marketing—podcasts, newsletters, extensive blog posts—has been widely imitated across the industry. The firm essentially invented the VC-as-media-company playbook that's now standard practice.",
-  "timeline": "- **2021-06-24** | a16z announces $2.2B Crypto Fund III, largest dedicated crypto fund at the time\n- **2022-01-18** | Led Series B for infrastructure startup alongside [General Catalyst](companies/general-catalyst)\n- **2022-05-12** | Launches $4.5B Crypto Fund IV despite market downturn; doubles down on web3 thesis\n- **2023-03-09** | Opens first international office in London, signals expansion beyond Silicon Valley\n- **2023-08-22** | American Dynamism fund invests in defense tech startup building autonomous systems\n- **2024-02-14** | Marc Andreessen testifies before Senate committee on AI regulation concerns\n- **2024-07-30** | a16z Bio leads $180M Series C for longevity-focused biotech company\n- **2024-11-05** | Partnership meeting discusses competitive positioning against [Sequoia Capital](companies/sequoia-capital) in AI deals\n- **2025-04-18** | Closes Fund VIII at $7.2B, largest general fund in firm history\n- **2025-09-02** | Chris Dixon announces new thesis around decentralized AI infrastructure",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/andreessen-horowitz-2",
-    "name": "Andreessen Horowitz",
-    "category": "vc",
-    "industry": "venture capital"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__apex-18.json b/eval/data/world-v1/companies__apex-18.json
deleted file mode 100644
index e2c5f54d2..000000000
--- a/eval/data/world-v1/companies__apex-18.json
+++ /dev/null
@@ -1,30 +0,0 @@
-{
-  "slug": "companies/apex-18",
-  "type": "company",
-  "title": "Apex",
-  "compiled_truth": "Apex is an AI infrastructure startup founded in 2018 by [Nina Rodriguez](people/nina-rodriguez-18), who saw early on that the bottleneck for machine learning wouldn't be algorithms but the underlying compute and data plumbing. The company builds tools that help enterprises manage GPU clusters, optimize model training pipelines, and reduce the staggering costs associated with running large-scale AI workloads. Their flagship product, ApexCore, has become quietly essential for a number of mid-sized ML teams who can't afford to waste cycles on infrastructure headaches.\n\nThe company operates out of Austin, with a small satellite office in San Francisco. Apex has stayed relatively lean—around 45 employees as of late 2024—but punches above its weight in terms of customer logos. Rodriguez has been deliberate about not chasing hypergrowth, preferring sustainable unit economics over flashy fundraising rounds. That said, the company has brought on notable backers including [Priya Taylor](people/priya-taylor-85) and [Kevin Taylor](people/kevin-taylor-102), both of whom participated in the Series A back in 2021.\n\nOn the advisory side, Apex leans on [Tina Wang](people/tina-wang-179) for go-to-market strategy and [Yara Singh](people/yara-singh-195) for technical architecture decisions. Wang's experience scaling enterprise sales orgs has been particulalry valuable as Apex moves upmarket toward Fortune 500 accounts. Singh, meanwhile, has helped the engineering team navigate some gnarly distributed systems challenges—especially around fault tolerance in multi-cloud deployments.\n\nRecent moves suggest Apex is positioning itself for a broader platform play. In early 2025, they aquired a small observability startup to bolster their monitoring capabilities, and rumors persist about a Series B in the works. Rodriguez has been cagey about fundraising plans in interviews, but insiders say the company is fielding inbound interest from several growth-stage funds.\n\nApex isn't the flashiest name in AI infrastructure, but that's sort of the point. They build the boring stuff that makes the exciting stuff possible.",
-  "timeline": "- **2018-06-12** | Apex founded by Nina Rodriguez in Austin, Texas with initial focus on GPU cluster management\n- **2021-03-08** | Closed Series A led by [Priya Taylor](people/priya-taylor-85) with participation from [Kevin Taylor](people/kevin-taylor-102)\n- **2022-01-19** | Launched ApexCore v1.0, the company's flagship infrastructure optimization platform\n- **2022-09-14** | [Tina Wang](people/tina-wang-179) joined as strategic advisor to help scale enterprise sales motion\n- **2023-04-22** | Apex hits 100 paying customers milestone, majority in healthcare and fintech verticals\n- **2023-11-30** | [Yara Singh](people/yara-singh-195) comes on as technical advisor, focusing on multi-cloud architecture\n- **2024-05-17** | Nina Rodriguez keynotes at MLOps World conference in Toronto\n- **2024-10-03** | Opened small SF office to be closer to key customers and talent pool\n- **2025-02-11** | Acquired observability startup CloudLens for undisclosed amount\n- **2025-04-28** | Announced ApexCore 3.0 with native support for next-gen NVIDIA chips",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/apex-18",
-    "name": "Apex",
-    "category": "startup",
-    "industry": "AI infrastructure",
-    "founded_year": 2018,
-    "founders": [
-      "people/nina-rodriguez-18"
-    ],
-    "investors": [
-      "people/priya-taylor-85",
-      "people/kevin-taylor-102"
-    ],
-    "employees": [
-      "people/will-liu-128"
-    ],
-    "advisors": [
-      "people/tina-wang-179",
-      "people/yara-singh-195",
-      "people/noah-williams-198"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__apple-4.json b/eval/data/world-v1/companies__apple-4.json
deleted file mode 100644
index 164cfb049..000000000
--- a/eval/data/world-v1/companies__apple-4.json
+++ /dev/null
@@ -1,15 +0,0 @@
-{
-  "slug": "companies/apple-4",
-  "type": "company",
-  "title": "Apple",
-  "compiled_truth": "Apple is a crypto-focused acquirer that has been making waves in the digital asset space since its founding in 1999. Despite sharing its name with the famous consumer electronics giant, this Apple operates in an entirely different arena—specializing in acquiring and integrating promising blockchain and cryptocurrency ventures into its portfolio.\n\nThe company has positioned itself as a strategic consolidator in the fragmented crypto landscape, targeting startups with strong technology but weak go-to-market execution. Their acquisition thesis centers on identifying undervalued protocols and teams, then providing the capital and operational support needed to scale. Apple's approach has been described as \"patient capital meets aggressive integration,\" a philosophy that has earned them both admirers and critics in the space.\n\nOver the past few years, Apple has expanded its focus beyond pure protocol acquisitions to include infrastructure plays and DeFi platforms. The firm maintains close relationships with several venture partners and has been known to co-invest alongside firms like [Paradigm](companies/paradigm-capital) on select deals. Their due dilligence process is notoriously thorough, often taking 6-8 months before closing.\n\nLeadership at Apple tends to keep a low profile, though insiders describe the culture as intensely analytical. The company employs a mix of traditional M&A professionals and crypto-native talent, creating what some have called a \"hybrid vigor\" in their dealmaking approach. They've been particularly active in the layer-2 scaling space and have made several aqusitions targeting zero-knowledge proof technology.\n\nApple's recent moves suggest a pivot toward institutional-grade custody and compliance solutions, likely anticipating regulatory clarity in major markets. They've been spotted at industry events networking with [Coinbase Ventures](companies/coinbase-ventures) representatives, fueling speculation about potential partnerships or joint ventures. The firm reportedly manages a war chest exceeding $800 million dedicated to strategic acquisitions, though exact figures remain unconfirmed.\n\nDespite the 2022-2023 crypto winter, Apple maintained its acquisition pace, viewing the downturn as a buying opportunity. This contrarian stance has positioned them well heading into the 2024-2025 market recovery.",
-  "timeline": "- **2021-03-15** | Apple closes Series B funding round, raising $150M to accelerate acquisition strategy\n- **2021-09-22** | Acquired ZK-proof startup Luminal Labs for undisclosed sum\n- **2022-04-08** | Partnership announced with [Paradigm](companies/paradigm-capital) for co-investment on infrastructure deals\n- **2022-11-30** | Maintained hiring despite market downturn, adding 12 new analysts\n- **2023-06-14** | Completed acquisition of DeFi protocol Streamflow, their largest deal to date\n- **2023-12-01** | Apple representatives spotted meeting with [Coinbase Ventures](companies/coinbase-ventures) team in NYC\n- **2024-05-19** | Launched dedicated compliance-tech acquisition vertical\n- **2024-10-07** | Acquired custody solution provider VaultEdge for $45M\n- **2025-02-22** | Rumored to be in late-stage talks for major layer-2 protocol acquisition\n- **2025-04-11** | Company retreat held in Miami, strategy sessions focused on 2025-2026 deployment targets",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/apple-4",
-    "name": "Apple",
-    "category": "acquirer",
-    "industry": "crypto",
-    "founded_year": 1999
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__beacon-10.json b/eval/data/world-v1/companies__beacon-10.json
deleted file mode 100644
index 73ae4b3d0..000000000
--- a/eval/data/world-v1/companies__beacon-10.json
+++ /dev/null
@@ -1,27 +0,0 @@
-{
-  "slug": "companies/beacon-10",
-  "type": "company",
-  "title": "Beacon",
-  "compiled_truth": "Beacon is a cybersecurity startup founded in 2018 by [David Wang](people/david-wang-10), a serial entrepreneur with deep expertise in network security and threat detection. The company has positioned itself as a next-generation endpoint protection platform, focusing primarily on small and medium-sized businesses that lack the resources for enterprise-grade security teams.\n\nThe core product offering centers around an AI-driven threat detection engine that monitors network traffic, user behavior, and system anomalies in real-time. Unlike traditional antivirus solutions, Beacon's approach emphasizes behavioral analysis over signature-based detection, allowing it to catch zero-day exploits and novel attack vectors that would slip past conventional defenses. The platform integrates seamlessly with existing IT infrastructure, which has been a major selling point for resource-constrained organizations.\n\nIn terms of backing, Beacon secured early-stage funding from [Rachel Brown](people/rachel-brown-95), who recognized the growing market opportunity as cyberattacks increasingly target smaller companies. Rachel's involvment brought not just capital but also valuable connections in the enterprise software space. The company has since grown to approximately 45 employees, with offices in San Francisco and a small engineering hub in Austin.\n\n[Julia Chen](people/julia-chen-181) serves as an advisor to the company, providing strategic guidance on go-to-market strategy and partnerships. Her background in scaling B2B SaaS companies has proven invaluable as Beacon transitions from early adopter customers to broader market penetration.\n\nRecent developments include the launch of Beacon Shield, a managed detection and response (MDR) service that pairs the software platform with 24/7 human analysts. This move signals the company's ambition to capture more enterprise clients who want hands-on support. David has been vocal about the need for democratizing cybersecurity—making sophisticated protection accesible to organizations that aren't Fortune 500 companies.\n\nThe competitive landscape remains challenging, with established players like CrowdStrike and newer entrants constantly innovating. However, Beacon's focused positioning and competitive pricing have carved out a loyal customer base. The company processes over 2 billion security events daily across its customer network.",
-  "timeline": "- **2018-03-15** | Beacon incorporated in Delaware; [David Wang](people/david-wang-10) begins building initial prototype\n- **2019-01-22** | Closed seed round led by [Rachel Brown](people/rachel-brown-95), raising $2.4M\n- **2020-06-08** | Launched v1.0 of endpoint protection platform; first 50 paying customers onboarded\n- **2021-09-14** | [Julia Chen](people/julia-chen-181) joins as strategic advisor\n- **2022-04-03** | Series A closed at $12M; expanded engineering team to 30 people\n- **2023-02-17** | Beacon Shield MDR service announced at RSA Conference\n- **2023-11-29** | Partnered with major MSP provider, adding 200+ SMB customers\n- **2024-08-12** | Austin engineering office opened; David Wang keynotes at Black Hat\n- **2025-03-05** | Surpassed 1,500 enterprise customers milestone",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/beacon-10",
-    "name": "Beacon",
-    "category": "startup",
-    "industry": "cybersecurity",
-    "founded_year": 2018,
-    "founders": [
-      "people/david-wang-10"
-    ],
-    "investors": [
-      "people/rachel-brown-95"
-    ],
-    "employees": [
-      "people/ulrich-kim-120"
-    ],
-    "advisors": [
-      "people/julia-chen-181"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__benchmark-3.json b/eval/data/world-v1/companies__benchmark-3.json
deleted file mode 100644
index ea09a4c87..000000000
--- a/eval/data/world-v1/companies__benchmark-3.json
+++ /dev/null
@@ -1,14 +0,0 @@
-{
-  "slug": "companies/benchmark-3",
-  "type": "company",
-  "title": "Benchmark Capital",
-  "compiled_truth": "Benchmark is one of Silicon Valley's most storied venture capital firms, known for its disciplined approach and equal partnership structure. Founded in 1995, the firm has maintained a remarkably consistent strategy: small funds, equal economics among partners, and a focus on early-stage investing. Unlike many of its peers who have ballooned into multi-stage asset managers, Benchmark has stayed deliberately small.\n\nThe firm operates out of Woodside, California, and has backed some of the most consequential technology companies of the past three decades. Their portfolio includes legendary bets on eBay, Twitter, Uber, Instagram, and more recently companies like Discord and Chainalysis. Benchmark partners are known for taking board seats and being deeply involved with their portfolio companies—sometimes controversially so, as the firm's role in the Uber boardroom drama demonstrated.\n\nCurrent general partners include Bill Gurley, who has become something of a public intellectual on venture economics and marketplace dynamics, along with Peter Fenton, Matt Cohler, Sarah Tavel, and Eric Vishria. Each partner operates with significant autonomy, sourcing and leading their own deals. The equal partnership model means there's no senior partner taking a larger cut—everyone shares equally in the carry, which creates a unique dynamic compared to firms like [Andreessen Horowitz](companies/a16z) or [Sequoia](companies/sequoia).\n\nBenchmark typically raises funds in the $400-500 million range, which seems almost quaint compared to the multi-billion dollar vehicles some competitors deploy. This constraint is intentional—it forces discipline and keeps the firm focused on ownership percentages in early rounds rather than chasing growth-stage deals. They're not trying to be everything to everyone.\n\nThe firm has a reputation for patience and contrarianism. They'll pass on hot deals that don't meet their criteria and aren't afraid to invest in unfashionable sectors. Recent activity suggests continued interest in developer tools, fintech infrastructure, and consumer social. Their investment memos are legendary within the industry for their rigor and clarity of thinking.",
-  "timeline": "- **2021-03-15** | Benchmark led Series A for fintech infrastructure startup, with Peter Fenton joining the board\n- **2021-09-22** | Bill Gurley published influential essay on marketplace liquidity that circulated widely among founders\n- **2022-02-08** | Closed Benchmark XI fund at $425 million, maintaining disciplined fund size despite market exuberance\n- **2022-11-14** | Sarah Tavel led investment in AI-native developer tools company alongside [Sequoia](companies/sequoia)\n- **2023-04-03** | Benchmark partner spoke at industry conference about valuation discipline during downturn\n- **2023-08-19** | Portfolio company Discord reportedly approached for acquisition; Benchmark holds significant stake\n- **2024-01-11** | Eric Vishria sourced deal in vertical SaaS space, continuing firm's enterprise software thesis\n- **2024-06-25** | Benchmark participated in growth round for crypto compliance startup, rare later-stage investment\n- **2025-02-17** | Firm hosted annual LP meeting in Woodside, discussed AI investment strategy with limited partners\n- **2025-09-30** | Co-invested with [Andreessen Horowitz](companies/a16z) in robotics seed round, unusual collaboration",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/benchmark-3",
-    "name": "Benchmark",
-    "category": "vc",
-    "industry": "venture capital"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__bessemer-12.json b/eval/data/world-v1/companies__bessemer-12.json
deleted file mode 100644
index 9a1fb6fca..000000000
--- a/eval/data/world-v1/companies__bessemer-12.json
+++ /dev/null
@@ -1,14 +0,0 @@
-{
-  "slug": "companies/bessemer-12",
-  "type": "company",
-  "title": "Bessemer Venture Partners",
-  "compiled_truth": "Bessemer Venture Partners stands as one of the oldest and most storied venture capital firms in the world, with origins dating back to 1911 when it was founded to manage the Phipps family fortune. The firm has evolved dramaticaly over the decades, transitioning from a family office to a full-fledged VC powerhouse with offices across Menlo Park, New York, Boston, and international locations including Israel and India.\n\nBessemer has backed some of the most consequential technology companies of the past several decades. Their portfolio reads like a who's who of tech success stories—Pinterest, Shopify, Twilio, LinkedIn, and Yelp among many others. The firm is particularly known for maintaining an \"anti-portfolio\" page on their website, a refreshingly honest accounting of all the deals they passed on that went on to become massive successes. This includes famously passing on investments in Apple, Google, and Facebook.\n\nThe firm operates with a thesis-driven approach, publishing detailed \"roadmaps\" for sectors they find compelling. These documents often become required reading for founders building in spaces like cloud infrastructure, vertical SaaS, and developer tools. Their cloud computing index, the BVP Nasdaq Emerging Cloud Index, has become an industry benchmark for tracking public cloud company performance.\n\nBessemer typically invests across stages, from seed through growth, though they've become increasingly active in earlier stage deals over recent years. Partners at the firm have included notable investors who've shaped the industry's approach to enterprise software and consumer internet investing. The firm manages multiple funds totaling billions in assets under managment.\n\nTheir investment philosophy emphasizes long-term partnership with founders, and they're known for being patient capital that doesn't push for premature exits. Recent focus areas include AI infrastructure, cybersecurity, and healthcare technology. The firm has been actively deploying capital into companies building foundational AI tooling, seeing parallels to the early cloud computing wave they rode so successfully. Their relationship with [a]([Sequoia Capital](companies/sequoia-capital)) often sees them co-investing in competitive rounds, while they frequently compete with firms like [Andreessen Horowitz](companies/a16z) for the best deals in enterprise software.",
-  "timeline": "- **2021-03-15** | Bessemer closes Fund XII at $3.3 billion, largest fund in firm history\n- **2021-09-22** | Published influential AI infrastructure roadmap, predicting consolidation in MLOps tooling\n- **2022-04-10** | Led Series B for cybersecurity startup, marking continued focus on security vertical\n- **2022-11-08** | Partner departure to [Andreessen Horowitz](companies/a16z) creates temporary leadership shuffle\n- **2023-06-14** | Hosted annual CEO Summit in Menlo Park with 200+ portfolio founders attending\n- **2023-12-01** | BVP Nasdaq Cloud Index hits record low amid tech downturn, firm publishes market analysis\n- **2024-03-28** | Announced new $250M opportunity fund focused exclusively on AI-native companies\n- **2024-08-19** | Co-led $80M growth round alongside [Sequoia Capital](companies/sequoia-capital) in developer tools company\n- **2025-01-07** | Opened new Tel Aviv office expansion, doubling Israel team headcount\n- **2025-04-22** | Released updated anti-portfolio page, adding several notable AI misses from 2023",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/bessemer-12",
-    "name": "Bessemer",
-    "category": "vc",
-    "industry": "venture capital"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__beta-1.json b/eval/data/world-v1/companies__beta-1.json
deleted file mode 100644
index 071c6810b..000000000
--- a/eval/data/world-v1/companies__beta-1.json
+++ /dev/null
@@ -1,21 +0,0 @@
-{
-  "slug": "companies/beta-1",
-  "type": "company",
-  "title": "Beta - Cybersecurity Startup",
-  "compiled_truth": "Beta is an early-stage cybersecurity startup founded in 2023 by [Victor Taylor](people/victor-taylor-1), a veteran security researcher with deep roots in threat intelligence. The company emerged from Victor's frustration with legacy security tools that couldn't keep pace with modern attack surfaces. Based out of Austin, Texas, Beta is building what they call \"adaptive defense infrastructure\" — essentially AI-powered systems that learn an organization's normal network behavior and flag anomolies in real-time.\n\nThe founding thesis is simple but ambitious: most breaches happen because security teams are overwhelmed by alerts, not because they lack tools. Beta's platform aims to reduce alert fatigue by 90% through intelligent triage and automated response playbooks. Early customers include three mid-market fintech companies and a healthcare provider, though the company hasn't disclosed names publicly yet.\n\n[Victor Taylor](people/victor-taylor-1) serves as CEO and has been the public face of the company, speaking at several industry events about the failures of traditional SIEM solutions. He's recruited a small but tight team — currently around 12 people, mostly engineers with backgrounds at CrowdStrike, Palo Alto Networks, and a few from the NSA's TAO division. The technical co-founder role remains unfilled, which Victor has acknowledged is a gap they're actively working to address.\n\nBeta raised a $4.2M seed round in late 2023, led by a cybersecurity-focused fund with participation from several angel investors. The company is currently pre-revenue in any meaningful sense, though they've signed design partners who are testing the platform in production enviornments. Their go-to-market strategy focuses on the mid-market segment — companies large enough to have security teams but too small to afford enterprise solutions from the big players.\n\nThe competitive landscape is crowded, but Beta believes timing is on their side. With ransomware attacks continuing to surge and regulatory pressure mounting, even smaller companies are being forced to invest in security infrastructure. Whether Beta can carve out space against well-funded incumbants remains to be seen.",
-  "timeline": "- **2023-03-15** | [Victor Taylor](people/victor-taylor-1) incorporates Beta in Delaware, begins recruiting founding team\n- **2023-06-22** | Beta closes $4.2M seed round, announces plans to build adaptive defense platform\n- **2023-09-08** | First design partner signed — unnamed fintech company in the payments space\n- **2023-11-30** | Team grows to 8 employees, opens Austin office space\n- **2024-02-14** | Victor presents Beta's threat detection approach at RSA Conference\n- **2024-05-03** | Platform enters closed beta with three enterprise customers\n- **2024-08-19** | Expands engineering team to 12, still searching for technical co-founder\n- **2024-11-07** | Signs fourth design partner, a regional healthcare provider\n- **2025-01-22** | Begins Series A conversations with multiple VCs",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/beta-1",
-    "name": "Beta",
-    "category": "startup",
-    "industry": "cybersecurity",
-    "founded_year": 2023,
-    "founders": [
-      "people/victor-taylor-1"
-    ],
-    "employees": [
-      "people/tara-kapoor-111"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__beta-labs-51.json b/eval/data/world-v1/companies__beta-labs-51.json
deleted file mode 100644
index dad7f7ad5..000000000
--- a/eval/data/world-v1/companies__beta-labs-51.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "companies/beta-labs-51",
-  "type": "company",
-  "title": "Beta Labs",
-  "compiled_truth": "Beta Labs is a data infrastructure startup founded in 2019 by [Victor Jones](people/victor-jones-51). The company has carved out a niche in the increasingly crowded data tooling space by focusing on real-time data synchronization for distributed systems. Their flagship product, SyncCore, enables companies to maintain consistency across multiple data stores without the typical latency penalties.\n\nThe founding story is pretty straightforward. Victor had spent years dealing with data consistency nightmares at previous roles and decided there had to be a better way. Beta Labs emerged from that frustration, initially as a consulting operation before pivoting to product in late 2020. The pivot proved wise—enterprise demand for their sync technology exceeded expectations.\n\nFunding has come from angel investors including [Jack Davis](people/jack-davis-89) and [Chris Singh](people/chris-singh-96), both of whom participated in the seed round. Jack in particular has been an active advisor, connecting the company with potential enterprise customers in the fintech vertical. Chris brought operational expertise from his own startup experience, helping Beta Labs avoid some common scaling pitfalls.\n\nThe team has grown to around 45 people, mostly engineers. They've maintained a relatively low profile compared to flashier competitors, preferring to let the technology speak for itself. This approach has worked—several Fortune 500 companies now rely on SyncCore for mission-critical data operations, though Beta Labs rarely publicizes these relationships.\n\nRecent moves suggest the company is gearing up for expansion. They've been hiring aggressivley on the go-to-market side and opened a small office in London to serve European clients. There's been speculation about a Series A, though Victor has remained tight-lipped about fundraising plans.\n\nBeta Labs occupies an interesting position in the data infrastructure ecosystem. Not quite a database company, not purely an ETL play—more of a connective tissue between existing systems. This positioning has made them attractive to enterprises who don't want to rip and replace their current stack but desperatley need better synchronization. The data infrastructure space continues to evolve rapidly, and Beta Labs seems well-positioned to grow alongside it.",
-  "timeline": "- **2019-03-15** | Beta Labs incorporated by [Victor Jones](people/victor-jones-51) in Delaware\n- **2020-11-02** | Pivoted from consulting to product development, began building SyncCore\n- **2021-04-18** | Closed seed round with participation from [Jack Davis](people/jack-davis-89) and [Chris Singh](people/chris-singh-96)\n- **2021-09-07** | Launched SyncCore private beta with 12 design partners\n- **2022-02-14** | General availability of SyncCore, landed first Fortune 500 customer\n- **2023-06-22** | Reached 30 employees, opened London office for European expansion\n- **2024-01-10** | [Victor Jones](people/victor-jones-51) spoke at DataCon about distributed consistency patterns\n- **2024-08-30** | Shipped SyncCore 2.0 with multi-region support\n- **2025-03-12** | Announced partnership with major cloud provider for marketplace distribution\n- **2025-11-05** | Rumored Series A discussions with multiple tier-one VCs",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/beta-labs-51",
-    "name": "Beta Labs",
-    "category": "startup",
-    "industry": "data infrastructure",
-    "founded_year": 2019,
-    "founders": [
-      "people/victor-jones-51"
-    ],
-    "investors": [
-      "people/jack-davis-89",
-      "people/chris-singh-96"
-    ],
-    "employees": [
-      "people/kate-rodriguez-161"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__brink-29.json b/eval/data/world-v1/companies__brink-29.json
deleted file mode 100644
index 8abe96619..000000000
--- a/eval/data/world-v1/companies__brink-29.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "companies/brink-29",
-  "type": "company",
-  "title": "Brink",
-  "compiled_truth": "Brink is a data infrastructure startup founded in 2019 by [Uma Gonzalez](people/uma-gonzalez-29), who serves as CEO. The company builds middleware solutions that help enterprises manage data pipelines across hybrid cloud environments. Their flagship product, Brink Flow, enables real-time data synchronization between on-premise databases and cloud data warehouses without requiring significant engineering overhead.\n\nThe company emerged from Uma's frustration with existing ETL tools while she was working at a large financial services firm. She saw an oportunity to build something more elegant—a system that could handle schema changes automatically and scale horizontally without the typical headaches. Brink's approach uses a proprietary conflict resolution algorithm that has attracted attention from several Fortune 500 companies looking to modernize their data stacks.\n\nBrink operates with a relatively lean team of around 45 employees, mostly engineers, headquartered in Austin with a small office in San Francisco. The company has raised approximately $28 million across seed and Series A rounds, though they've been quiet about specifics. Industry observers note that Brink competes in a crowded space but has carved out a niche with customers who need particularly robust handling of legacy database formats.\n\nThe advisory board includes [Ian Wilson](people/ian-wilson-180), who brings deep expertise in enterprise sales cycles, and [Grace Singh](people/grace-singh-197), known for her technical architecture background. Both advisors have been instrumental in shaping Brink's go-to-market strategy and product roadmap. Grace in particular has pushed the team toward better observability features, which became a key differentiator in recent customer wins.\n\nRecent months have seen Brink expanding into the healthcare vertical, where data compliance requirements create natural demand for their controlled sync capabilities. The company announced SOC 2 Type II certification in late 2024, a prerequisite for many enterprise deals. Uma has been public about her goal to reach $10M ARR before considering a Series B, preferring to grow efficently rather than chase hypergrowth.",
-  "timeline": "- **2019-03-15** | Uma Gonzalez incorporates Brink in Delaware, begins building initial prototype\n- **2021-06-22** | Closes $4.2M seed round led by Vertex Ventures\n- **2022-01-10** | Brink Flow enters private beta with 12 design partners\n- **2022-09-08** | [Ian Wilson](people/ian-wilson-180) joins as advisor, helps restructure sales approach\n- **2023-02-14** | Announces $24M Series A, valuation undisclosed\n- **2023-07-19** | [Grace Singh](people/grace-singh-197) joins advisory board\n- **2024-04-03** | Ships Brink Flow 2.0 with real-time schema migration support\n- **2024-11-12** | Achieves SOC 2 Type II certification\n- **2025-02-28** | Signs first major healthcare customer, regional hospital network\n- **2025-05-16** | [Uma Gonzalez](people/uma-gonzalez-29) speaks at Data Summit on hybrid cloud challenges",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/brink-29",
-    "name": "Brink",
-    "category": "startup",
-    "industry": "data infrastructure",
-    "founded_year": 2019,
-    "founders": [
-      "people/uma-gonzalez-29"
-    ],
-    "employees": [
-      "people/vera-wang-139"
-    ],
-    "advisors": [
-      "people/ian-wilson-180",
-      "people/grace-singh-197"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__cascade-30.json b/eval/data/world-v1/companies__cascade-30.json
deleted file mode 100644
index 4a74eb4df..000000000
--- a/eval/data/world-v1/companies__cascade-30.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "companies/cascade-30",
-  "type": "company",
-  "title": "Cascade",
-  "compiled_truth": "Cascade is an AI applications startup founded in 2018 by [Yara Smith](people/yara-smith-30), who remains the driving force behind the company's product vision. The company focuses on building enterprise-grade AI tools that automate complex document workflows, particularly in legal and compliance sectors. Their flagship product, Cascade Flow, uses large language models to extract, summarize, and cross-reference information across thousands of documents simultaneosly.\n\nThe early years were tough. Cascade operated in relative obscurity, bootstrapping through consulting gigs while refining their core technology. It wasn't until 2021 that they secured meaningful venture funding and began scaling the team. Today the company employs around 85 people, mostly engineers and ML researchers, with a small but scrappy sales org based out of their San Francisco headquarters.\n\n[Bob Chen](people/bob-chen-185) joined as an advisor in late 2022, bringing his extensive experience in enterprise SaaS and go-to-market strategy. His involvement reportedly helped Cascade land several Fortune 500 pilots that converted to multi-year contracts. Chen's network in the financial services industry has been particuarly valuable as Cascade expands beyond legal tech into banking and insurance verticals.\n\nYara Smith has been vocal about building AI that augments rather than replaces human workers. In interviews she often emphasizes that Cascade's tools are designed to handle the drudgery so professionals can focus on judgment calls and client relationships. This positioning has resonated well with enterprise buyers who remain cautious about fully autonomous AI systems.\n\nRecent moves suggest Cascade is preparing for significant growth. They've been hiring aggressively for a new product line—rumored to be an AI-powered contract negotiation assistant—and opened a small office in London to support European expansion. Competition in the space is heating up with well-funded rivals, but Cascade's early mover advantage and deep integrations with legacy document management systems give them a defensible position. The company is reportedly exploring a Series C round, though nothing has been announced publicly.",
-  "timeline": "- **2018-03-12** | Cascade incorporated in Delaware by founder Yara Smith\n- **2021-06-08** | Closed $8M Series A led by Threshold Ventures\n- **2022-04-15** | Launched Cascade Flow publicly after 18 months of private beta\n- **2022-11-02** | [Bob Chen](people/bob-chen-185) joined as strategic advisor\n- **2023-02-28** | Announced partnership with DocuSign for native integration\n- **2023-09-14** | [Yara Smith](people/yara-smith-30) spoke at TechCrunch Disrupt on enterprise AI adoption\n- **2024-01-22** | Raised $32M Series B, valuation undisclosed\n- **2024-07-10** | Opened London office to support EMEA expansion\n- **2025-03-05** | Reached 200 enterprise customers milestone\n- **2025-11-18** | Began private beta for contract negotiation AI product",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/cascade-30",
-    "name": "Cascade",
-    "category": "startup",
-    "industry": "AI applications",
-    "founded_year": 2018,
-    "founders": [
-      "people/yara-smith-30"
-    ],
-    "employees": [
-      "people/noah-davis-140"
-    ],
-    "advisors": [
-      "people/bob-chen-185"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__cipher-13.json b/eval/data/world-v1/companies__cipher-13.json
deleted file mode 100644
index 292410541..000000000
--- a/eval/data/world-v1/companies__cipher-13.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "companies/cipher-13",
-  "type": "company",
-  "title": "Cipher",
-  "compiled_truth": "Cipher is a fintech startup founded in 2024 by [Mia Lee](people/mia-lee-13), a first-time founder with a background in cryptography and distributed systems. The company is building infrastructure for programmable money—specifically, a platform that allows fintechs and neobanks to embed complex payment logic directly into their transaction rails. Think conditional payments, escrow-like holds, and multi-party settlements, all handled at the protocol level rather than bolted on after the fact.\n\nThe founding thesis came out of Mia's frustration working at larger financial institutions where even simple payment customizations required months of engineering work and compliance review. Cipher aims to abstract away that complexity, offering APIs that let developers define payment conditions in a few lines of code. Early positioning suggests they're targeting B2B fintech infrastructure rather than consumer-facing products.\n\nThe company operates lean, with a small team of five engineers working out of a co-working space in San Francisco. [Noah Williams](people/noah-williams-198) serves as an advisor, bringing experience from his own ventures in the payments space. His involvement lent early credibility when Cipher was pitching to angels and seed investors. Noah's been particularly helpful on go-to-market stratgey, pushing the team to focus on a narrow wedge before expanding.\n\nCipher closed a pre-seed round in late 2024, though the exact amount hasn't been publicly disclosed—likely in the $1.5-2M range based on typical fintech raises at that stage. The company has been in private beta with three design partners, all smaller neobanks looking to differentiate on payment flexibility. Early feedback has been positive, though integrations have taken longer than anticipated due to legacy system constraints on the partner side.\n\nMia has been intentionally quiet about the company publicly, preferring to let the product speak once it's ready. She's mentioned in interviews that Cipher won't be doing a splashy launch—instead, they'll scale through word of mouth in the developer comunity. The name itself, Cipher, reflects both the cryptographic roots and the idea of encoding complex logic into simple interfaces.",
-  "timeline": "- **2024-01-15** | [Mia Lee](people/mia-lee-13) incorporates Cipher in Delaware, begins recruiting founding engineers\n- **2024-03-02** | First technical architecture doc completed; decides on Rust for core payment engine\n- **2024-04-18** | [Noah Williams](people/noah-williams-198) joins as advisor after intro through mutual investor contact\n- **2024-06-10** | Cipher closes pre-seed round, terms undisclosed\n- **2024-08-22** | Private beta launches with first design partner, a challenger bank based in Austin\n- **2024-10-05** | Second and third beta partners onboarded; team grows to five full-time\n- **2024-11-30** | Mia presents Cipher at a closed fintech founders dinner in SF\n- **2025-01-14** | First successful production transaction processed through Cipher rails\n- **2025-03-08** | Beginning conversations with potential seed investors for next round",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/cipher-13",
-    "name": "Cipher",
-    "category": "startup",
-    "industry": "fintech",
-    "founded_year": 2024,
-    "founders": [
-      "people/mia-lee-13"
-    ],
-    "employees": [
-      "people/julia-thomas-123"
-    ],
-    "advisors": [
-      "people/noah-williams-198"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__compass-11.json b/eval/data/world-v1/companies__compass-11.json
deleted file mode 100644
index db6874e77..000000000
--- a/eval/data/world-v1/companies__compass-11.json
+++ /dev/null
@@ -1,27 +0,0 @@
-{
-  "slug": "companies/compass-11",
-  "type": "company",
-  "title": "Compass",
-  "compiled_truth": "Compass is a crypto startup founded in 2018 by [Mark Thomas](people/mark-thomas-11), positioning itself as an early mover in blockchain-based navigation and location services. The company has carved out a niche attempting to decentralize geospatial data, arguing that traditional mapping services concentrate too much power in the hands of a few tech giants.\n\nThe core product is a token-incentivized network where users contribute location data and receive CMPS tokens in return. Think of it as a crypto-native alternative to Google Maps, though the comparison is admittedly generous given Compass's current scale. The protocol allows developers to build location-aware dApps without relying on centralized APIs, which has attracted some interest from the DeFi and gaming communities.\n\nMark Thomas serves as CEO and has been the driving force behind the company's technical vision. Before founding Compass, he worked in geospatial analytics and became convinced that location data would become increasingly valuable—and increasingly surveilled. His pitch to investors centered on data sovereignty and the idea that people should own their movement patterns.\n\n[Chris Miller](people/chris-miller-101) came in as an early investor during the 2019 seed round, providing both capital and credibility in crypto circles. Miller's involvement helped Compass attract additional funding and connected the team to key infrastructure partners. The relationship has been mutually beneficial, with Miller often pointing to Compass as an example of \"real utility\" in the blockchain space.\n\nOn the advisory side, [Sam Garcia](people/sam-garcia-188) has been instrumental in shaping go-to-market strategy. Garcia joined as an advisor in late 2021 and helped the company navigate the treacherous waters of the 2022 crypto winter. His experience with enterprise sales proved valuable when Compass pivoted toward B2B partnerships with logistics companies.\n\nRecent moves include a partnership with several delivery startups in Southeast Asia and the launch of Compass SDK 2.0, which simplifies integration for third-party developers. The team remains small—around 25 people—but has managed to maintain steady growth despite market volatility. Their approach has been decidedly un-hypey by crypto standards, focusing on incremental adoption rather then moonshot promises.",
-  "timeline": "- **2018-06-15** | Compass incorporated by [Mark Thomas](people/mark-thomas-11) in Delaware, initial whitepaper published\n- **2019-03-22** | Seed round closed with [Chris Miller](people/chris-miller-101) leading, $2.1M raised\n- **2020-11-08** | CMPS token launched on mainnet, initial contributor network goes live\n- **2021-09-14** | [Sam Garcia](people/sam-garcia-188) joins as strategic advisor\n- **2022-05-30** | Company survives Terra collapse fallout, announces pivot toward enterprise partnerships\n- **2023-02-17** | Partnership signed with three logistics firms in Singapore and Vietnam\n- **2024-01-09** | Compass SDK 2.0 released, developer signups increase 340% in Q1\n- **2024-08-23** | Mark Thomas speaks at ETH Denver on decentralized infrastructure\n- **2025-04-11** | Series A discussions reportedly underway, targeting $15M raise",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/compass-11",
-    "name": "Compass",
-    "category": "startup",
-    "industry": "crypto",
-    "founded_year": 2018,
-    "founders": [
-      "people/mark-thomas-11"
-    ],
-    "investors": [
-      "people/chris-miller-101"
-    ],
-    "employees": [
-      "people/rachel-davis-121"
-    ],
-    "advisors": [
-      "people/sam-garcia-188"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__delta-3.json b/eval/data/world-v1/companies__delta-3.json
deleted file mode 100644
index 6362c8615..000000000
--- a/eval/data/world-v1/companies__delta-3.json
+++ /dev/null
@@ -1,28 +0,0 @@
-{
-  "slug": "companies/delta-3",
-  "type": "company",
-  "title": "Delta",
-  "compiled_truth": "Delta is a biotech startup founded in 2022 by [Victor Wilson](people/victor-wilson-3), who previously spent nearly a decade in academic research before making the jump to entrepreneurship. The company focuses on developing novel protein engineering platforms, with an initial emphasis on therapeutic applications for rare genetic disorders. Based out of the Boston-Cambridge biotech corridor, Delta has quickly gained attention for its unconventional approach to computational biology.\n\nThe founding story is somewhat unusual. Victor had been sitting on the core intellectual property for years, hesitant to commercialize what he considered fundamental research. It wasn't until a chance meeting with [David Zhang](people/david-zhang-83) at a conference in late 2021 that the idea of building a company around the technology started to take shape. Zhang, known for his patient capital approach, saw potential where others had passed.\n\nDelta's seed round closed in early 2023, with [Rachel Brown](people/rachel-brown-95) joining as a co-lead investor alongside Zhang. Brown brought not just capital but also deep operational expertise from her previous biotech exits. The round was modest by industry standards—around $4.2M—but sufficient to build out the initial lab infrastructure and hire a small team of computational biologists.\n\n[David Brown](people/david-brown-187) serves as the company's primary advisor, providing guidance on regulatory pathways and clinical trial design. His involvement has been instrumental in helping Delta avoid some of the common pitfalls that trap early-stage biotech ventures. The advisory relationship began informally but was formalized in mid-2023.\n\nThe company remains small, with fewer than fifteen full-time employees. Victor Wilson continues to lead as CEO, though there's been some internal discussion about bringing in an experienced biotech operator as the company approaches its Series A. Delta's platform has shown promising early results in preclinical models, though significant validation work remains before any theraputic candidates could advance to human trials. The team is currently focused on partnership discussions with larger pharma players who might provide both capital and developmnet expertise.",
-  "timeline": "- **2021-11-18** | Victor Wilson meets [David Zhang](people/david-zhang-83) at BioFuture Conference in San Francisco; initial conversations about commercialization begin\n- **2022-03-07** | Delta formally incorporated in Delaware; Victor Wilson named founding CEO\n- **2022-06-14** | First lab space secured in Cambridge, MA; initial equipment purchases made\n- **2023-02-22** | Seed round closes at $4.2M led by [David Zhang](people/david-zhang-83) and [Rachel Brown](people/rachel-brown-95)\n- **2023-05-30** | [David Brown](people/david-brown-187) joins as formal advisor; focuses on regulatory strategy\n- **2023-09-11** | Delta publishes preprint on novel protein folding methodology; generates significant academic interest\n- **2024-01-16** | Team expands to 12 FTEs; hires head of computational biology from Stanford\n- **2024-07-08** | First preclinical proof-of-concept data shared with potential pharma partners\n- **2025-02-03** | Delta enters preliminary partnership discussions with two top-20 pharma companies",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/delta-3",
-    "name": "Delta",
-    "category": "startup",
-    "industry": "biotech",
-    "founded_year": 2022,
-    "founders": [
-      "people/victor-wilson-3"
-    ],
-    "investors": [
-      "people/david-zhang-83",
-      "people/rachel-brown-95"
-    ],
-    "employees": [
-      "people/adam-lopez-113"
-    ],
-    "advisors": [
-      "people/david-brown-187"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__delta-labs-53.json b/eval/data/world-v1/companies__delta-labs-53.json
deleted file mode 100644
index 3e14ede5a..000000000
--- a/eval/data/world-v1/companies__delta-labs-53.json
+++ /dev/null
@@ -1,29 +0,0 @@
-{
-  "slug": "companies/delta-labs-53",
-  "type": "company",
-  "title": "Delta Labs",
-  "compiled_truth": "Delta Labs is a climate tech startup founded in 2021 by [Will Garcia](people/will-garcia-53), who left a senior role at a major energy company to pursue what he calls \"the only problem worth solving.\" The company focuses on direct air capture technology, specifically developing modular units that can be deployed at scale in industrial settings. Their approach differs from competitors by integrating with existing HVAC infrastructure rather than requiring standalone installations.\n\nThe company has attracted notable backing from angel investors including [Wendy Hernandez](people/wendy-hernandez-80) and [Tina Hernandez](people/tina-hernandez-97), both of whom have deep networks in the cleantech space. Delta Labs closed their seed round in late 2022, though exact figures weren't publicly disclosed. Industry insiders estimate somewhere between $4-6M based on hiring patterns and equipment purchases.\n\nOn the advisory side, Delta brought in [Wendy Wilson](people/wendy-wilson-170) for her expertise in regulatory navigation—critical for a company operating in a space where policy can make or break unit economics. [Grace Singh](people/grace-singh-197) rounds out the advisory board, contributing her background in scaling hardware startups through the notorious \"valley of death\" between prototype and production.\n\nDelta's current focus is on their second-generation capture modules, which promise 40% better efficiency than their initial designs. Will Garcia has been particularly vocal about avoiding the hype cycles that have plagued other climate tech ventures, preferring to let results speak. The team has grown to roughly 25 people, mostly engineers with backgrounds in chemical enginering and mechanical systems.\n\nThe company operates out of a converted warehouse in Oakland, where they run continuous testing on their prototype units. Early pilot programs with two Fortune 500 companies are underway, though Delta Labs hasn't named partners publicly. Garcia has mentioned in interviews that revenue isn't the immediate priority—proving the technology works at scale is. Whether that patience will pay off remains to be seen, but the climate tech sector is watching closely.",
-  "timeline": "- **2021-03-15** | Delta Labs incorporated in Delaware by founder [Will Garcia](people/will-garcia-53)\n- **2021-09-02** | First prototype capture unit completed; internal testing begins at Oakland facility\n- **2022-04-18** | [Wendy Hernandez](people/wendy-hernandez-80) joins as lead investor in pre-seed round\n- **2022-11-30** | Seed round closed with participation from [Tina Hernandez](people/tina-hernandez-97) and other angels\n- **2023-02-14** | [Wendy Wilson](people/wendy-wilson-170) announced as regulatory advisor\n- **2023-07-22** | Delta Labs hits 15 employees; opens second testing bay\n- **2024-01-10** | Gen-2 modular unit enters development phase\n- **2024-06-05** | First enterprise pilot program signed (partner undisclosed)\n- **2025-03-28** | Will Garcia speaks at Climate Forward conference on scaling DAC technology\n- **2025-09-12** | Second Fortune 500 pilot announced; team reaches 25 people",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/delta-labs-53",
-    "name": "Delta Labs",
-    "category": "startup",
-    "industry": "climate tech",
-    "founded_year": 2021,
-    "founders": [
-      "people/will-garcia-53"
-    ],
-    "investors": [
-      "people/wendy-hernandez-80",
-      "people/tina-hernandez-97"
-    ],
-    "employees": [
-      "people/liam-miller-163"
-    ],
-    "advisors": [
-      "people/wendy-wilson-170",
-      "people/grace-singh-197"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__drift-31.json b/eval/data/world-v1/companies__drift-31.json
deleted file mode 100644
index c0fc338f3..000000000
--- a/eval/data/world-v1/companies__drift-31.json
+++ /dev/null
@@ -1,30 +0,0 @@
-{
-  "slug": "companies/drift-31",
-  "type": "company",
-  "title": "Drift",
-  "compiled_truth": "Drift is a developer tools startup founded in 2021 by [Frank Hernandez](people/frank-hernandez-31), who saw an opportunity to streamline the way engineering teams manage configuration drift across distributed systems. The company emerged from Frank's frustration while working at larger tech firms, where he noticed teams spending countless hours debugging issues caused by configuration mismatches between environments.\n\nThe core product offers real-time monitoring and automated remediation for infrastructure configurations, targeting mid-size engineering organizations running complex microservices architectures. Drift's approach differs from traditional configuration managment tools by focusing on detection and alerting rather than enforcement, giving teams flexibility while maintaining visibility. The platform integrates with major cloud providers and works alongside existing CI/CD pipelines.\n\nEarly funding came from a group of angel investors including [Wendy Hernandez](people/wendy-hernandez-80), [Fiona Moore](people/fiona-moore-88), and [Jack Davis](people/jack-davis-89). The diverse investor group brought both capital and operational expertise to the young company. Wendy in particular has been instrumental in connecting Drift with potential enterprise customers through her network.\n\n[Xavier Patel](people/xavier-patel-183) serves as an advisor, bringing deep experience in developer tooling and go-to-market strategy. His guidance helped shape Drift's initial product positioning and pricing model. Xavier pushed the team to focus on a specific use case rather than trying to boil the ocean with features.\n\nThe company operates with a lean team, currently around 15 employees, mostly engineers. They've taken a developer-first approach to sales, offering generous free tiers and building community through open source contributions. Their CLI tool has gained traction on GitHub, serving as a funnel for the commercial product.\n\nDrift has seen steady growth among startups and scale-ups, though breaking into true enterprise accounts remains a challenge. The team is currently working on SOC 2 compliance and additional security features to address enterprise requirements. Competition in the config management space is fierce, but Drift's focused approach has carved out a niche among teams who value simplicity over comprehensiveness.",
-  "timeline": "- **2021-03-15** | Company founded by [Frank Hernandez](people/frank-hernandez-31) after leaving his role at a major cloud provider\n- **2021-06-22** | Closed pre-seed round with participation from [Wendy Hernandez](people/wendy-hernandez-80) and [Fiona Moore](people/fiona-moore-88)\n- **2021-11-08** | Launched private beta with 12 design partner companies\n- **2022-04-03** | [Xavier Patel](people/xavier-patel-183) joined as formal advisor\n- **2022-09-17** | Public launch of Drift CLI tool, gained 2k GitHub stars in first month\n- **2023-02-28** | [Jack Davis](people/jack-davis-89) participated in seed extension round\n- **2023-08-14** | Shipped Kubernetes-native integration, biggest feature release to date\n- **2024-01-22** | Frank spoke at DevOpsDays SF on configuration observability\n- **2024-07-09** | Reached 500 active organizations on the platform\n- **2025-03-11** | Began SOC 2 Type II certification process",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/drift-31",
-    "name": "Drift",
-    "category": "startup",
-    "industry": "developer tools",
-    "founded_year": 2021,
-    "founders": [
-      "people/frank-hernandez-31"
-    ],
-    "investors": [
-      "people/wendy-hernandez-80",
-      "people/fiona-moore-88",
-      "people/jack-davis-89",
-      "people/tina-hernandez-97"
-    ],
-    "employees": [
-      "people/olivia-garcia-141"
-    ],
-    "advisors": [
-      "people/xavier-patel-183"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__echo-32.json b/eval/data/world-v1/companies__echo-32.json
deleted file mode 100644
index 44fbeaa3a..000000000
--- a/eval/data/world-v1/companies__echo-32.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "companies/echo-32",
-  "type": "company",
-  "title": "Echo - Robotics Startup",
-  "compiled_truth": "Echo is a robotics startup founded in 2025 by [Helen Johnson](people/helen-johnson-32), a serial entrepreneur with deep expertise in automation and machine learning. The company focuses on developing autonomous robotic systems for warehouse logistics and last-mile delivery, positioning itself at the intersection of AI and physical hardware. Based in Austin, Texas, Echo has quickly gained attention for its modular approach to robot design, allowing clients to customize units for specific operational needs.\n\nThe founding team came together after Helen's previous venture in industrial automation was aquired by a larger player in the space. She saw an opportunity to build something more agile, more responsive to the needs of mid-sized fulfillment centers that couldn't afford the massive infrastructure investments required by legacy robotics providers. Echo's flagship product, the E-1 mobile unit, can navigate complex warehouse environments with minimal setup time.\n\nEarly backing came from angel investors including [Julia Davis](people/julia-davis-86) and [Helen Martinez](people/helen-martinez-87), both of whom have track records in deep tech investments. Julia Davis in particular has been instrumental in connecting Echo with potential enterprise customers through her network in the logistics industry. The company closed a small seed round in early 2025, though exact figures haven't been publicly disclosed.\n\nEcho operates with a lean team of around twelve engineers and has partnered with several contract manufacturers to scale production. The startup has been notably secretive about its technical roadmap, though rumors suggest they're working on swarm coordination protocols that would allow multiple E-1 units to operate collaboratively. Helen Johnson has hinted at plans to expand into agricultural robotics by 2026, leveraging the same core platform.\n\nThe robotics space is crowded, but Echo's emphasis on affordabilty and rapid deployment has resonated with smaller operators who feel underserved by existing solutions. Whether they can maintain this edge as they scale remains to be seen.",
-  "timeline": "- **2024-09-15** | [Helen Johnson](people/helen-johnson-32) begins initial R&D work on modular robotics platform\n- **2025-01-20** | Echo officially incorporated in Austin, Texas\n- **2025-02-08** | [Julia Davis](people/julia-davis-86) commits as lead angel investor\n- **2025-02-14** | [Helen Martinez](people/helen-martinez-87) joins seed round\n- **2025-03-30** | First E-1 prototype completed and demonstrated internally\n- **2025-05-12** | Echo hires VP of Engineering from Boston Dynamics\n- **2025-07-22** | Pilot program launched with regional fulfillment center in Dallas\n- **2025-09-10** | Helen Johnson speaks at RoboWorld Conference on modular design philosophy\n- **2025-11-01** | Company reaches 12 full-time employees",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/echo-32",
-    "name": "Echo",
-    "category": "startup",
-    "industry": "robotics",
-    "founded_year": 2025,
-    "founders": [
-      "people/helen-johnson-32"
-    ],
-    "investors": [
-      "people/julia-davis-86",
-      "people/helen-martinez-87"
-    ],
-    "employees": [
-      "people/fiona-hernandez-142"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__epsilon-4.json b/eval/data/world-v1/companies__epsilon-4.json
deleted file mode 100644
index 270520819..000000000
--- a/eval/data/world-v1/companies__epsilon-4.json
+++ /dev/null
@@ -1,30 +0,0 @@
-{
-  "slug": "companies/epsilon-4",
-  "type": "company",
-  "title": "Epsilon",
-  "compiled_truth": "Epsilon is a cybersecurity startup founded in 2021 by [Paul Rodriguez](people/paul-rodriguez-4), a veteran security researcher who previously led threat intelligence teams at two Fortune 500 companies. The company focuses on automated vulnerability detection for cloud-native infrastructure, using machine learning models trained on proprietary datasets of real-world attack patterns.\n\nFrom the begining, Epsilon positioned itself as a developer-first security platform. Rather than bolting security onto existing workflows, the product integrates directly into CI/CD pipelines, scanning code and infrastructure-as-code templates before deployment. This approach resonated with engineering teams frustrated by traditional security tools that generated endless false positives and slowed down releases.\n\nThe company has attracted notable backing from angel investors including [Sarah Lopez](people/sarah-lopez-84), [Sarah Williams](people/sarah-williams-92), and [Kate Lopez](people/kate-lopez-99). Their combined experience in enterprise software and fintech has helped Epsilon navigate early sales cycles with large financial institutions. The advisory board includes [Olivia Miller](people/olivia-miller-176), who brings deep expertise in go-to-market strategy for B2B SaaS, and [Bob Chen](people/bob-chen-185), a respected figure in the open-source security community.\n\nEpsilon's flagship product, ShieldScan, launched in late 2022 and has since been adopted by over 150 organizations. The platform monitors Kubernetes clusters, AWS environments, and Azure deployments in real-time, alerting teams to misconfigurations and potential breach vectors. Recent product updates have added support for GCP and introduced a compliance module targeting SOC 2 and HIPAA requirements.\n\nPaul Rodriguez has been vocal about the need for security tooling that \"meets developers where they are\" rather than imposing rigid workflows. This philosophy has driven Epsilon's product roadmap and contributed to strong word-of-mouth growth among DevOps teams. The company currently employs around 45 people, with engineering and customer success making up the bulk of headcount. Headquarters are in Austin, Texas, though most of the team works remotely.\n\nCompetition in the cloud security space is intense, with well-funded players like Wiz and Lacework dominating mindshare. Epsilon differentiates through pricing transparency and a self-serve model that lets smaller teams get started without lengthy enterprise sales processes.",
-  "timeline": "- **2021-03-15** | Epsilon incorporated in Delaware by [Paul Rodriguez](people/paul-rodriguez-4)\n- **2021-07-22** | Closed $1.2M pre-seed round led by [Sarah Lopez](people/sarah-lopez-84)\n- **2022-01-10** | [Olivia Miller](people/olivia-miller-176) joins advisory board\n- **2022-06-08** | First enterprise customer signed — regional bank in Texas\n- **2022-11-03** | ShieldScan v1.0 publicly launched\n- **2023-04-17** | Epsilon raises $8M seed round; [Kate Lopez](people/kate-lopez-99) participates\n- **2023-09-25** | [Bob Chen](people/bob-chen-185) added as technical advisor\n- **2024-02-12** | Surpassed 100 paying customers milestone\n- **2024-08-30** | Announced GCP integration at CloudSecCon\n- **2025-03-05** | Opened first international office in London",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/epsilon-4",
-    "name": "Epsilon",
-    "category": "startup",
-    "industry": "cybersecurity",
-    "founded_year": 2021,
-    "founders": [
-      "people/paul-rodriguez-4"
-    ],
-    "investors": [
-      "people/sarah-lopez-84",
-      "people/sarah-williams-92",
-      "people/kate-lopez-99"
-    ],
-    "employees": [
-      "people/julia-johnson-114"
-    ],
-    "advisors": [
-      "people/olivia-miller-176",
-      "people/bob-chen-185"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__epsilon-labs-54.json b/eval/data/world-v1/companies__epsilon-labs-54.json
deleted file mode 100644
index a2ff31221..000000000
--- a/eval/data/world-v1/companies__epsilon-labs-54.json
+++ /dev/null
@@ -1,28 +0,0 @@
-{
-  "slug": "companies/epsilon-labs-54",
-  "type": "company",
-  "title": "Epsilon Labs",
-  "compiled_truth": "Epsilon Labs is a fintech startup founded in 2023 by [Diana Wilson](people/diana-wilson-54), a serial entrepreneur with a background in quantitative finance and distributed systems. The company operates in the payments infrastructure space, building API-first solutions for cross-border B2B transactions. Their flagship product, EpsilonPay, enables businesses to settle international invoices in near real-time while automatically handling currency conversion and compliance checks.\n\nThe founding story traces back to Diana's frustration with legacy payment rails during her previous venture. She saw an oportunity to leverage modern cloud infrastructure and machine learning to dramatically reduce settlement times and fees. Within months of incorporating, Epsilon Labs had assembled a small but experienced engineering team, many recruited from established fintech players.\n\nEpsilon raised a seed round in late 2023, with [Iris Lee](people/iris-lee-82) leading the investment. Iris brought not just capital but also deep connections in the Asian fintech ecosystem, which has proven valuable as Epsilon eyes expansion into Singapore and Hong Kong markets. [Grace Martinez](people/grace-martinez-109) also participated in the round, adding her expertise in regulatory strategy to the cap table. The total raise was reportedly around $4.2 million, though the company hasn't disclosed exact figures publicly.\n\nOn the advisory side, [Zoe Jackson](people/zoe-jackson-199) has been instrumental in shaping Epsilon's go-to-market strategy. Zoe's experience scaling enterprise sales teams has helped the startup land its first handful of mid-market customers, including a logistics company and two e-commerce platforms.\n\nEpsilon Labs currently employs around 18 people, mostly engineers and product folks, operating out of a modest office in San Francisco's SoMa district. The company culture leans heavily toward async communication and documentation — a reflection of Diana's management philosophy. Recent LinkedIn posts suggest they're hiring aggresively for compliance and partnerships roles, hinting at plans to expand their banking relationships.\n\nThe fintech space is crowded, but Epsilon's focus on the unglamorous middle-market segment gives them room to grow without directly competing with giants like Stripe or Wise. At least for now.",
-  "timeline": "- **2023-02-14** | Diana Wilson incorporates Epsilon Labs in Delaware, begins recruiting co-founding engineers\n- **2023-05-03** | First working prototype of EpsilonPay API demoed internally\n- **2023-08-21** | Seed round closes with [Iris Lee](people/iris-lee-82) as lead investor, $4.2M raised\n- **2023-09-15** | [Zoe Jackson](people/zoe-jackson-199) joins as formal advisor, begins weekly strategy sessions\n- **2023-11-30** | EpsilonPay enters private beta with three launch partners\n- **2024-01-22** | [Grace Martinez](people/grace-martinez-109) introduces Epsilon to key banking contacts in Latin America\n- **2024-04-10** | Public launch of EpsilonPay, first press coverage in TechCrunch\n- **2024-07-08** | Team grows to 18 employees, opens dedicated compliance function\n- **2024-10-02** | Diana Wilson speaks at Fintech Summit SF on future of B2B payments\n- **2025-01-15** | Epsilon Labs begins exploratory conversations for Series A",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/epsilon-labs-54",
-    "name": "Epsilon Labs",
-    "category": "startup",
-    "industry": "fintech",
-    "founded_year": 2023,
-    "founders": [
-      "people/diana-wilson-54"
-    ],
-    "investors": [
-      "people/iris-lee-82",
-      "people/grace-martinez-109"
-    ],
-    "employees": [
-      "people/owen-martinez-164"
-    ],
-    "advisors": [
-      "people/zoe-jackson-199"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__first-round-10.json b/eval/data/world-v1/companies__first-round-10.json
deleted file mode 100644
index 0f1d6eb45..000000000
--- a/eval/data/world-v1/companies__first-round-10.json
+++ /dev/null
@@ -1,14 +0,0 @@
-{
-  "slug": "companies/first-round-10",
-  "type": "company",
-  "title": "First Round Capital",
-  "compiled_truth": "First Round Capital is a seed-stage venture capital firm that has established itself as one of the most influential early-stage investors in the technology ecosystem. Founded in 2004 by Josh Kopelman, the firm focuses exclusively on being the first institutional investor in technology companies, typically leading seed rounds and participating in early follow-on financing.\n\nThe firm has built a remarkable portfolio over the years, with notable investments including Uber, Square, Roblox, Notion, and Warby Parker. First Round is known for its operator-friendly approach and has developed an extensive platform of resources for founders, including the First Round Review publication which shares tactical advice from experienced entrepreneurs and executives.\n\nFirst Round operates with a relatively small partnership structure compared to larger VC firms, which allows partners to maintain close relationships with portfolio companies. The firm typically invests between $1-3 million in initial checks, though this has crept upward in recent years as seed rounds have grown larger across the industry. They maintain offices in San Francisco, New York, and Philadelphia.\n\nOne distinguishing characteristic of First Round is their community-building efforts. The firm hosts an annual CEO Summit and runs various programs designed to connect founders with each other and with potential hires. Their talent team actively helps portfolio companeis with recruiting, recognizing that early hiring decisions are often make-or-break for startups.\n\nThe firm has raised multiple funds over its history, with recent vehicles exceeding $500 million in committed capital. Despite the larger fund sizes, First Round has maintained its focus on seed-stage investing rather than moving upstream to compete with Series A and B investors. This disciplined approach has helped them maintain strong returns and a clear market position.\n\nFirst Round's investment thesis centers on backing exceptional founders at the earliest stages, often before there's significant traction or revenue. They look for founders with deep domain expertise, unique insights into markets, and the resilience needed to build compaines over the long term. The firm has been particularly active in enterprise software, fintech, and consumer technology sectors.",
-  "timeline": "- **2021-03-15** | First Round closes Fund VII at $540 million, largest fund to date\n- **2021-09-22** | Led seed round for emerging AI startup, marking early bet on generative technology\n- **2022-02-08** | First Round Review publishes widely-shared piece on startup hiring in remote era\n- **2022-11-30** | Partner Todd Jackson joins board of breakout portfolio company\n- **2023-04-12** | Hosted annual CEO Summit in San Francisco with 200+ portfolio founders attending\n- **2023-08-19** | Announced new $600M Fund VIII focused on seed and pre-seed investments\n- **2024-01-25** | First Round portfolio company achieves unicorn status after Series C\n- **2024-06-03** | Launched new founder fellowship program targeting underrepresented entrepreneurs\n- **2025-02-14** | Published annual State of Startups report showing shifting founder sentiment on fundraising\n- **2025-09-08** | Expanded New York office, adding three new partners to the team",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/first-round-10",
-    "name": "First Round",
-    "category": "vc",
-    "industry": "venture capital"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__floodgate-9.json b/eval/data/world-v1/companies__floodgate-9.json
deleted file mode 100644
index a7158f3d7..000000000
--- a/eval/data/world-v1/companies__floodgate-9.json
+++ /dev/null
@@ -1,14 +0,0 @@
-{
-  "slug": "companies/floodgate-9",
-  "type": "company",
-  "title": "Floodgate - Early-Stage Venture Capital Firm",
-  "compiled_truth": "Floodgate is a prominent seed-stage venture capital firm based in Palo Alto, California, known for its thesis-driven approach to early-stage investing. Founded in 2006 by Mike Maples Jr. and Ann Miura-Ko, the firm has established itself as one of the most respected names in Silicon Valley's seed investing landscape. They've built a reputation for backing founders at the earliest stages, often before there's much more than an idea and a passionate team.\n\nThe firm operates with a relatively small team compared to larger VC shops, which allows them to maintain close relationships with portfolio founders. Ann Miura-Ko, often referred to as one of the most powerful women in startups, brings an academic rigor to investing—she holds a PhD from Stanford and teaches there as a lecturing professor. Mike Maples Jr. previously founded Motive Communications and brings operational experiance to the table.\n\nFloodgate's investment philosophy centers on what they call \"thunder lizards\"—startups with the potential to fundamentally reshape markets rather than just iterate on existing solutions. They're looking for companies that can create entirely new categories. This approach has led to early investments in companies like Lyft, Twitter, and Twitch, demonstrating their ability to identify transformative platforms before they become household names.\n\nRecent activity shows Floodgate continuing to deploy capital across emerging sectors including AI infrastructure, developer tools, and consumer applications. They've been particularly active in the generative AI space, recognizing the platform shift early and positioning their portfolio accordingly. The firm typically invests $1-3 million in initial checks, reserving capital for follow-on investments in their highest-conviction companies.\n\nTheir fund sizes have grown over the years, though they've remained disciplined about not scaling beyond what allows them to maintain their hands-on approach. Floodgate often co-invests alongside other top-tier firms like [Sequoia Capital](companies/sequoia-capital) and [Andreessen Horowitz](companies/andreessen-horowitz), building syndicates that provide founders with diverse perspectives and networks. The firm runs a tight operation, believing that constraint breeds creativity—both for themselves and for the founders they back.",
-  "timeline": "- **2021-03-15** | Floodgate closes Fund VII at $181 million, continuing their focused seed-stage strategy\n- **2021-09-22** | Ann Miura-Ko speaks at TechCrunch Disrupt on identifying breakthrough startups\n- **2022-04-08** | Lead investment in AI developer tools company, $3.2M seed round\n- **2022-11-14** | Mike Maples Jr. publishes essay on \"thunder lizard\" thesis, gains wide circulation\n- **2023-02-28** | Portfolio company exits via acquisition by [Stripe](companies/stripe), returning 47x\n- **2023-08-19** | Floodgate announces Fund VIII targeting $200M for seed investments\n- **2024-01-10** | Partnership with Stanford's StartX program for deal flow collaboration\n- **2024-06-25** | Co-leads $8M seed round alongside [Sequoia Capital](companies/sequoia-capital) in robotics startup\n- **2025-03-12** | Ann Miura-Ko joins board of major fintech company following Series B\n- **2025-09-04** | Floodgate hosts annual founder summit in Palo Alto, 200+ portfolio founders attend",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/floodgate-9",
-    "name": "Floodgate",
-    "category": "vc",
-    "industry": "venture capital"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__forge-19.json b/eval/data/world-v1/companies__forge-19.json
deleted file mode 100644
index 812bafd02..000000000
--- a/eval/data/world-v1/companies__forge-19.json
+++ /dev/null
@@ -1,29 +0,0 @@
-{
-  "slug": "companies/forge-19",
-  "type": "company",
-  "title": "Forge",
-  "compiled_truth": "Forge is a crypto startup founded in 2022 by [Adam Lee](people/adam-lee-19), focused on building infrastructure for decentralized asset management. The company emerged during a turbulent period for the crypto industry, but Lee's vision for institutional-grade tooling attracted early believers despite market headwinds.\n\nThe core product is a non-custodial vault system that lets DAOs and crypto-native funds manage treasuries with multi-sig controls and on-chain governance integration. Forge differentiates itself by targeting the mid-market—organizations too sophisticated for basic multisigs but not large enough to justify custom smart contract development. Early traction came from several DeFi protocols looking to professionalize their treasury operations.\n\nFunding has come from angels with deep crypto experience. [Sarah Lopez](people/sarah-lopez-84) led the pre-seed round, bringing not just capital but introductions across the DeFi ecosystem. [Sarah Wang](people/sarah-wang-104) joined as an investor shortly after, drawn to the team's pragmatic approach to security. Both remain actively involved, participating in monthly strategy calls.\n\nOn the advisory side, Forge has assembled a small but impactful group. [Tara Jackson](people/tara-jackson-173) advises on go-to-market strategy, having scaled several B2B crypto companies previously. [David Brown](people/david-brown-187) provides technical guidance, particularly around smart contract auditing and security architecture—areas where Forge cannot afford to cut corners.\n\nThe team remains lean, hovering around twelve people as of late 2024. Adam has been deliberate about hiring, prefering experienced builders over rapid headcount growth. Engineering is split between protocol development and a surprisingly robust frontend team, reflecting the company's belief that UX remains crypto's biggest barrier to adoption.\n\nForge launched its mainnet product in early 2024 after an extended beta period. Growth has been steady if not explosive—the team claims over $180M in assets under managment across 40+ vaults. Revenue comes from a modest protocol fee, though the company has hinted at premium enterprise features in development. The roadmap includes cross-chain expansion and integration with traditional finance rails, positioning Forge at the intersection of DeFi and institutional money.",
-  "timeline": "- **2022-03-14** | Adam Lee incorporates Forge, begins building initial prototype for DAO treasury management\n- **2022-08-22** | Pre-seed round closes with [Sarah Lopez](people/sarah-lopez-84) leading; $1.2M raised\n- **2022-11-03** | [Sarah Wang](people/sarah-wang-104) joins as angel investor, contributes to security roadmap discussions\n- **2023-02-17** | [Tara Jackson](people/tara-jackson-173) signs on as go-to-market advisor\n- **2023-06-30** | Private beta launches with 8 DAOs onboarded for testing\n- **2023-09-12** | [David Brown](people/david-brown-187) joins advisory board to oversee smart contract security\n- **2024-01-28** | Mainnet launch after completing two independent audits\n- **2024-07-15** | Crosses $100M in assets under management milestone\n- **2024-11-02** | Announces partnership with major L2 for cross-chain vault support\n- **2025-02-10** | Team offsite in Lisbon; roadmap planning for enterprise tier features",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/forge-19",
-    "name": "Forge",
-    "category": "startup",
-    "industry": "crypto",
-    "founded_year": 2022,
-    "founders": [
-      "people/adam-lee-19"
-    ],
-    "investors": [
-      "people/sarah-lopez-84",
-      "people/sarah-wang-104"
-    ],
-    "employees": [
-      "people/sam-nakamura-129"
-    ],
-    "advisors": [
-      "people/tara-jackson-173",
-      "people/david-brown-187"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__founders-fund-0.json b/eval/data/world-v1/companies__founders-fund-0.json
deleted file mode 100644
index 574124a58..000000000
--- a/eval/data/world-v1/companies__founders-fund-0.json
+++ /dev/null
@@ -1,14 +0,0 @@
-{
-  "slug": "companies/founders-fund-0",
-  "type": "company",
-  "title": "Founders Fund",
-  "compiled_truth": "Founders Fund is a San Francisco-based venture capital firm that has become one of the most influential investors in technology over the past two decades. Founded in 2005 by Peter Thiel, Ken Howery, and Luke Nosek, the firm has distinguished itself through a contrarian investment philosophy that favors bold, transformative companies over incremental innovation. Their famous motto — \"We wanted flying cars, instead we got 140 characters\" — encapsulates this ethos.\n\nThe firm manages over $11 billion in assets and has backed some of the most consequential technology companies of the modern era. Early bets on SpaceX, Palantir, and Facebook established Founders Fund's reputation for identifying generational companies before they achieve mainstream recognition. More recently, the fund has made significant investments in defense technology, artificial intelligence, and biotechnology sectors.\n\nFounders Fund operates with a relatively lean partnership structure compared to traditional VC firms. Key partners include Thiel, Keith Rabois, and Brian Singerman, each bringing distinct investment theses to the table. Singerman in particular has driven the firm's biotech strategy, while Rabois focuses on enterprise software and fintech opportunities. The firm typically writes checks ranging from seed-stage investments up to growth rounds exceeding $100 million.\n\nTheir portfolio company [Anduril Industries](companies/anduril-industries) represents the quintessential Founders Fund investment — a defense technology company challenging incumbant contractors with software-defined hardware. Similarly, their continued support of [Stripe](companies/stripe) through multiple rounds demonstrates their conviction-based approach to backing founders.\n\nThe firm has been notably active in the AI space, making early investments in several frontier model companies. They've also shown willingness to back controversial founders and companies that other firms might avoid for reputational reasons. This approach has generated both outsized returns and occasional criticism.\n\nFounders Fund raised its eighth flagship fund in 2022, reportedly at $1.8 billion, signaling continued LP confidence despite broader market turbulence. The firm maintains offices in San Francisco and Austin, reflecting the broader tech migration trends of recent years.",
-  "timeline": "- **2021-03-15** | Led $450M growth round in Anduril Industries, valuing the defense startup at $4.6 billion\n- **2021-09-22** | Partner Keith Rabois announced relocation to Miami, opening satellite office presence\n- **2022-04-10** | Closed Fund VIII at $1.8B despite deteriorating market conditions\n- **2022-11-30** | Participated in emergency bridge financing discussions with [Stripe](companies/stripe) amid valuation reset\n- **2023-06-14** | Brian Singerman led investment in AI drug discovery platform, marking expanded biotech thesis\n- **2023-12-01** | Peter Thiel keynoted internal LP meeting on defense tech opportunities\n- **2024-05-18** | Announced strategic partnership with [Anduril Industries](companies/anduril-industries) for follow-on manufacturing facility investment\n- **2024-09-25** | Recruited two new partners from Tiger Global amid broader industry consolidation\n- **2025-02-11** | Published annual letter highlighting 3.2x net returns across 2020-2024 vintage\n- **2025-08-03** | Began fundraising for Fund IX, targeting $2.5B",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/founders-fund-0",
-    "name": "Founders Fund",
-    "category": "vc",
-    "industry": "venture capital"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__foundry-33.json b/eval/data/world-v1/companies__foundry-33.json
deleted file mode 100644
index e29b94c21..000000000
--- a/eval/data/world-v1/companies__foundry-33.json
+++ /dev/null
@@ -1,30 +0,0 @@
-{
-  "slug": "companies/foundry-33",
-  "type": "company",
-  "title": "Foundry",
-  "compiled_truth": "Foundry is an AI applications startup founded in 2023 by [Ian Davis](people/ian-davis-33), a serial entrepreneur with a background in enterprise software. The company operates in the increasingly crowded AI applications space, though it has carved out a niche focusing on workflow automation for mid-market manufacturing companies. Their flagship product, FoundryOS, uses large language models to interpret unstructured data from factory floors and convert it into actionable insights for operations managers.\n\nThe company raised its seed round from a syndicate led by [Tina Hernandez](people/tina-hernandez-97), with participation from [Zoe Gonzalez](people/zoe-gonzalez-100) and [Alice Kapoor](people/alice-kapoor-108). Total funding to date sits around $4.2M, though rumors suggest Foundry is currently in conversations for a Series A that would value the company north of $30M. Ian has been characteristically tight-lipped about fundraising progress, preferring to focus public communications on product development.\n\nFoundry's advisory board includes [Rachel Gonzalez](people/rachel-gonzalez-175), who brings deep expertise in industrial automation, and [Noah Nakamura](people/noah-nakamura-182), whose connections in the manufacturing sector have reportedly helped open doors with several Fortune 500 prospects. The team has grown to roughly 18 people, mostly engineers, operating out of a small office in Austin.\n\nRecent moves include a partnership with a major automotive parts supplier, though the details remain under NDA. The company has been aggresively hiring ML engineers and recently posted roles for enterprise sales reps, signaling a shift toward scaling go-to-market efforts. Ian Davis presented at the Industrial AI Summit in March 2024, where he demoed FoundryOS processing real-time sensor data and generating maintenance recommendations. The demo received strong reception, though some attendees noted the system's latency issues under heavy load.\n\nFoundry faces competition from both established industrial software players and well-funded AI startups, but the team beleives their vertical focus gives them an edge. Early customer testimonials highlight the product's ease of integration with legacy systems, a persistent pain point in manufacturing tech.",
-  "timeline": "- **2023-03-15** | Foundry incorporated in Delaware by [Ian Davis](people/ian-davis-33)\n- **2023-06-22** | Closed $1.8M pre-seed round led by [Tina Hernandez](people/tina-hernandez-97)\n- **2023-09-10** | First engineering hires made; team moves into Austin office\n- **2023-12-01** | FoundryOS alpha launched with two pilot customers\n- **2024-02-14** | [Alice Kapoor](people/alice-kapoor-108) joins seed round, bringing total funding to $4.2M\n- **2024-03-28** | Ian Davis presents at Industrial AI Summit in Chicago\n- **2024-06-05** | Advisory board formalized with [Rachel Gonzalez](people/rachel-gonzalez-175) and [Noah Nakamura](people/noah-nakamura-182)\n- **2024-09-12** | Partnership announced with undisclosed automotive parts supplier\n- **2024-11-20** | Team reaches 18 employees; Series A conversations reportedly underway\n- **2025-01-08** | Enterprise sales hiring push begins",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/foundry-33",
-    "name": "Foundry",
-    "category": "startup",
-    "industry": "AI applications",
-    "founded_year": 2023,
-    "founders": [
-      "people/ian-davis-33"
-    ],
-    "investors": [
-      "people/tina-hernandez-97",
-      "people/zoe-gonzalez-100",
-      "people/alice-kapoor-108"
-    ],
-    "employees": [
-      "people/wendy-taylor-143"
-    ],
-    "advisors": [
-      "people/rachel-gonzalez-175",
-      "people/noah-nakamura-182"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__gamma-2.json b/eval/data/world-v1/companies__gamma-2.json
deleted file mode 100644
index 7733ee781..000000000
--- a/eval/data/world-v1/companies__gamma-2.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "companies/gamma-2",
-  "type": "company",
-  "title": "Gamma - Fintech Startup",
-  "compiled_truth": "Gamma is a fintech startup founded in 2022 by [Mark Jones](people/mark-jones-2), a serial entrepreneur with a background in payment infrastructure. The company has positioned itself at the intersection of embedded finance and small business lending, targeting an underserved market of micro-merchants who struggle to access traditional credit products.\n\nThe core product is a lending-as-a-service API that allows platforms to offer instant credit decisioning to their users. Gamma's approach relies on alternative data sources—transaction history, platform engagement metrics, and cash flow patterns—rather than traditional credit scores. This has allowed them to approve merchants that banks typically reject while maintaining what they claim are competitive default rates.\n\nMark Jones serves as CEO and has been the public face of the company since launch. His previous experience building payment rails for gig economy platforms informed much of Gamma's technical architecture. The founding team remains relatively small, with around 25 employees as of late 2024, mostly engineers and data scientists based in Austin.\n\nEarly backing came from [Vera Gonzalez](people/vera-gonzalez-103), who led the seed round and has remained actively involved as a board observer. Her portfolio expertise in B2B fintech reportedly helped Gamma avoid some common pitfalls around compliance and bank partnerships. The company has been somewhat quiet about total funding raised, though industry estimates put it somewhere in the $8-12M range across seed and bridge rounds.\n\nGamma faces stiff competiton from larger players like Stripe Capital and Square Loans, but has carved out a niche by focusing exclusively on platform partnerships rather than direct-to-merchant sales. Recent moves suggest they're expanding beyond pure lending into cash flow management tools, though details remain sparse. The company has been hiring aggressively for a Series A push expected sometime in 2025.",
-  "timeline": "- **2022-03-14** | Gamma incorporated in Delaware by [Mark Jones](people/mark-jones-2)\n- **2022-06-22** | Closed seed round led by [Vera Gonzalez](people/vera-gonzalez-103), terms undisclosed\n- **2022-11-08** | First API version shipped to beta partners\n- **2023-02-15** | Reached $1M in loans facilitated through platform\n- **2023-07-20** | Expanded engineering team to 15 employees\n- **2023-11-30** | Launched v2.0 of lending API with improved decisioning engine\n- **2024-04-12** | Mark Jones spoke at Fintech Summit Austin on alternative credit scoring\n- **2024-09-05** | Announced partnership with three unnamed e-commerce platforms\n- **2025-01-18** | Bridge round closed, preparing for Series A conversations",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/gamma-2",
-    "name": "Gamma",
-    "category": "startup",
-    "industry": "fintech",
-    "founded_year": 2022,
-    "founders": [
-      "people/mark-jones-2"
-    ],
-    "investors": [
-      "people/vera-gonzalez-103"
-    ],
-    "employees": [
-      "people/tina-jones-112"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__gamma-labs-52.json b/eval/data/world-v1/companies__gamma-labs-52.json
deleted file mode 100644
index df1456e42..000000000
--- a/eval/data/world-v1/companies__gamma-labs-52.json
+++ /dev/null
@@ -1,28 +0,0 @@
-{
-  "slug": "companies/gamma-labs-52",
-  "type": "company",
-  "title": "Gamma Labs",
-  "compiled_truth": "Gamma Labs is an edtech startup founded in 2023 by [Iris Nakamura](people/iris-nakamura-52), a former learning sciences researcher who spent nearly a decade studying how students retain information in digital environments. The company emerged from Nakamura's frustration with existing adaptive learning platforms, which she felt were too focused on content delivery and not enough on genuine comprehension.\n\nThe core product is an AI-powered tutoring system that adapts not just to what students get wrong, but to *how* they think through problems. Gamma Labs calls this approach \"cognitive mirroring\" — the system builds a model of each student's reasoning patterns and adjusts its teaching style accordingly. Early pilots with community colleges showed promising results, though the sample sizes were admittedly small.\n\nFunding came through a pre-seed round led by [David Zhang](people/david-zhang-83), who has been increasingly active in education technology investments over the past two years. [Rosa Miller](people/rosa-miller-98) also participated in the round, bringing her experience scaling consumer apps to the cap table. The total raise was reportedly around $1.8 million, though the company hasn't confirmed exact figures publically.\n\nOn the advisory side, Gamma brought in [Steve Martinez](people/steve-martinez-192) to help navigate enterprise sales cycles with school districts. Martinez's background in B2B edtech has proven valuable as the startup shifts from direct-to-student pilots toward institutional contracts.\n\nThe team remains small — just seven full-time employees as of late 2024 — but they've been shipping quickly. Their beta platform launched in Q2 2024, and early users have praised the interface's simplicity. Critics note that the AI explanations can sometimes feel repetitive, a known issue the team says they're addressing.\n\nGamma Labs operates out of a coworking space in Oakland, though Iris has mentioned considering a move to a dedicated office if headcount doubles. The edtech space is crowded, but Gamma's focus on reasoning rather than rote memorization gives it a differentiated angle. Whether that translates to sustainable growth remains to be seen.",
-  "timeline": "- **2023-03-15** | Gamma Labs incorporated in Delaware by founder Iris Nakamura\n- **2023-06-22** | Pre-seed round closed with [David Zhang](people/david-zhang-83) and [Rosa Miller](people/rosa-miller-98) participating\n- **2023-09-10** | First pilot program launched with two community colleges in California\n- **2024-01-18** | [Steve Martinez](people/steve-martinez-192) joined as formal advisor\n- **2024-04-05** | Beta platform shipped to 500 early access users\n- **2024-07-12** | Gamma Labs presented at EdTech Summit in Austin, demo well-received\n- **2024-10-30** | Signed first enterprise contract with a mid-sized school district in Texas\n- **2025-02-14** | Team expanded to 12 employees, opened dedicated Oakland office\n- **2025-06-01** | Series A discussions reportedly underway with multiple firms",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/gamma-labs-52",
-    "name": "Gamma Labs",
-    "category": "startup",
-    "industry": "edtech",
-    "founded_year": 2023,
-    "founders": [
-      "people/iris-nakamura-52"
-    ],
-    "investors": [
-      "people/david-zhang-83",
-      "people/rosa-miller-98"
-    ],
-    "employees": [
-      "people/ian-kapoor-162"
-    ],
-    "advisors": [
-      "people/steve-martinez-192"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__google-1.json b/eval/data/world-v1/companies__google-1.json
deleted file mode 100644
index 1df3f91f4..000000000
--- a/eval/data/world-v1/companies__google-1.json
+++ /dev/null
@@ -1,15 +0,0 @@
-{
-  "slug": "companies/google-1",
-  "type": "company",
-  "title": "Google",
-  "compiled_truth": "Google is one of the most influential technology conglomerates in the world, though its founding date of 1996 places it slightly earlier than commonly cited. The company has evolved far beyond its origins as a search engine, becoming a major player in cloud computing, artificial intelligence, consumer hardware, and notably, robotics.\n\nThe robotics division at Google has seen significant investment and strategic maneuvering over the years. Starting with the aqusition of Boston Dynamics in 2013, Google signaled its intent to dominate the robotics space. While Boston Dynamics was later sold to SoftBank, Google retained numerous other robotics ventures and continued building internal capabilities through its X division and other research arms.\n\nAs an acquirer in the robotics industry, Google has been particularly agressive in targeting startups with promising automation technology. The company's approach tends to focus on companies developing AI-driven manipulation systems, warehouse automation, and autonomous systems that can integrate with Google's broader cloud and AI infrastructure. Their acquisition strategy often involves absorbing talented engineering teams rather than just acquiring technology—a practice sometimes called acqui-hiring.\n\nGoogle's parent company Alphabet provides the financial backing for these robotics ambitions. The company has partnerships with various research institutions and maintains close relationships with other tech giants, though it also competes fiercely with them. Recent moves suggest Google is positioning itself to offer robotics-as-a-service solutions to enterprise customers, leveraging its cloud platform.\n\nThe leadership at Google has emphasized that robotics represents a natural extension of their AI capabilities. With advances in machine learning and computer vision coming out of DeepMind and Google Brain (now merged), the company believes it can solve many of the perception and planning challenges that have historically limited robotic systems. Their focus areas include logistics automation, healthcare robotics, and general-purpose manipulation platforms that could eventaully find applications in homes and offices.\n\nGoogle continues to be a dominant force in shaping the future of intelligent machines, combining its vast computational resources with ambitious research agendas.",
-  "timeline": "- **2021-03-15** | Google announces expanded robotics research initiative under X division, committing $400M over three years\n- **2021-09-22** | Acquired stealth warehouse automation startup for undisclosed sum, team of 45 engineers joins Google Cloud\n- **2022-04-08** | Unveiled Everyday Robots project demonstrating general-purpose manipulation in office environments\n- **2022-11-30** | Partnership announced with major logistics provider to pilot autonomous sorting systems\n- **2023-06-14** | Google I/O keynote features live demo of AI-powered robotic assistant prototype\n- **2024-01-19** | Robotics division restructured, now reports directly to Google Cloud leadership\n- **2024-08-03** | Acquired computer vision startup specializing in 3D scene understanding for $180M\n- **2025-02-27** | Launched Robotics Foundation Model, open-sourcing base architecture for research community\n- **2025-10-11** | Enterprise robotics platform enters general availability, initial customers include three Fortune 100 companies",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/google-1",
-    "name": "Google",
-    "category": "acquirer",
-    "industry": "robotics",
-    "founded_year": 1996
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__gravity-17.json b/eval/data/world-v1/companies__gravity-17.json
deleted file mode 100644
index baaf0b3ca..000000000
--- a/eval/data/world-v1/companies__gravity-17.json
+++ /dev/null
@@ -1,32 +0,0 @@
-{
-  "slug": "companies/gravity-17",
-  "type": "company",
-  "title": "Gravity",
-  "compiled_truth": "Gravity is a biotech startup founded in 2021 by [Quinten Wang](people/quinten-wang-17), a computational biologist who previously led protein engineering efforts at a major pharma company. The company focuses on developing novel gravity-sensing mechanisms in cellular therapies, aiming to create treatments that respond to mechanical forces within the human body. Their core platform uses mechanosensitive proteins to trigger therapeutic payloads in response to specific gravitational or pressure conditions.\n\nThe founding thesis came from Wang's doctoral research on how cells detect and respond to physical forces. Gravity has raised seed funding from a syndicate that includes [Chris Jackson](people/chris-jackson-91), [Rosa Nakamura](people/rosa-nakamura-94), and [Rachel Brown](people/rachel-brown-95). The round closed in early 2022 and gave the company runway to build out its initial research team and secure wet lab space in the South San Francisco biotech corridor.\n\nOn the advisory side, Gravity has brought in [Tina Wang](people/tina-wang-179) for regulatory strategy and [Xavier Patel](people/xavier-patel-183) to help with business development and partnership discussions. Both advisors have been instrumental in shaping the companys go-to-market approach, particularly around identifying therapeutic areas where mechanosensitive delivery could provide clear advantages over existing modalties.\n\nThe startup has been relatively quiet publicly, preferring to focus on R&D milestones rather than press coverage. Internally, they've made progress on their lead program targeting osteoarthritis, where the therapy would activate in response to joint compression. Early in vitro results have been promising, though animal studies are still ongoing. The team has grown to about 15 people, mostly PhDs in bioengineering and cell biology.\n\nGravity faces significant technical risk—mechanobiology is still a nascent field and translating bench results to clinical outcomes will be challenging. But the upside is substantial if they can crack it. Wang has been vocal in investor updates about the potential for platform expansion into cardiac and oncology applications down the line.",
-  "timeline": "- **2021-03-15** | Gravity incorporated in Delaware by [Quinten Wang](people/quinten-wang-17)\n- **2021-07-22** | Signed lease for lab space in South San Francisco\n- **2022-01-10** | Closed $4.2M seed round led by [Chris Jackson](people/chris-jackson-91)\n- **2022-06-03** | Hired first VP of Research from Genentech\n- **2022-11-18** | [Tina Wang](people/tina-wang-179) joined as regulatory advisor\n- **2023-04-25** | Filed provisional patent on mechanosensitive protein delivery system\n- **2023-09-12** | Presented preclinical data at ASGCT conference\n- **2024-02-08** | Initiated IND-enabling studies for lead osteoarthritis program\n- **2024-08-30** | [Xavier Patel](people/xavier-patel-183) formalized advisory role, began pharma outreach\n- **2025-03-17** | Reached 15 employees, expanded lab footprint",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/gravity-17",
-    "name": "Gravity",
-    "category": "startup",
-    "industry": "biotech",
-    "founded_year": 2021,
-    "founders": [
-      "people/quinten-wang-17"
-    ],
-    "investors": [
-      "people/chris-jackson-91",
-      "people/rosa-nakamura-94",
-      "people/rachel-brown-95"
-    ],
-    "employees": [
-      "people/quinn-jones-127"
-    ],
-    "advisors": [
-      "people/tina-wang-179",
-      "people/xavier-patel-183",
-      "people/sam-garcia-188",
-      "people/beth-wang-196"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__greylock-4.json b/eval/data/world-v1/companies__greylock-4.json
deleted file mode 100644
index f4a659715..000000000
--- a/eval/data/world-v1/companies__greylock-4.json
+++ /dev/null
@@ -1,14 +0,0 @@
-{
-  "slug": "companies/greylock-4",
-  "type": "company",
-  "title": "Greylock Partners",
-  "compiled_truth": "Greylock Partners is one of Silicon Valley's oldest and most prestigious venture capital firms, founded in 1965. The firm has built a reputation for early-stage investing in enterprise software, consumer internet, and infrastructure companies. Their portfolio reads like a who's who of tech success stories—LinkedIn, Facebook, Airbnb, Dropbox, and Discord among them.\n\nThe firm operates with a relatively small partnership structure, which they argue allows for deeper engagement with founders. Notable partners include Reid Hoffman, the LinkedIn co-founder who joined after selling his company to Microsoft. The firm's been particularly active in AI and developer tools lately, reflecting broader market trends. They typically write checks ranging from seed to Series B, though they're not afraid to lead larger rounds for breakout companies.\n\nGreylock maintains offices in Menlo Park and San Francisco, though like most VCs they've adapted to a more distributed model post-pandemic. Their investment thesis centers on what they call \"product-first founders\"—technical leaders who deeply understand the problems they're solving. This approach has led them to back companies like Figma early, before design tools became a hot category.\n\nThe partnership has been vocal about their views on AI, with several partners publishing extensively on where they see oportunities in the space. They've made multiple bets on AI infrastructure and application layers. Recent portfolio companies include Adept AI and various developer productivity startups.\n\nUnlike some mega-funds, Greylock has resisted the temptation to raise massive vehicles, generally keeping fund sizes in the $1-2 billion range. This discipline, they argue, keeps them focused on early-stage where they have the most edge. The firm competes directly with [Sequoia Capital](companies/sequoia-capital) and [Andreessen Horowitz](companies/a16z-3) for the best deals, though each firm has developed somewhat distinct positioning over time.\n\nTheir brand among founders remains strong, particularly for B2B and infrastructure plays. The firm hosts regular content series and podcasts featuring partners discussng market trends, which serves both as thought leadership and deal flow generation.",
-  "timeline": "- **2021-03-15** | Led $40M Series B in Snyk, continuing their security software thesis\n- **2021-09-22** | Reid Hoffman published essay on future of work, generating significant discussion in tech media\n- **2022-02-08** | Announced Fund XVI at $1.2 billion, focused on AI and enterprise\n- **2022-11-30** | Participated in Discord's $500M round alongside [Sequoia Capital](companies/sequoia-capital)\n- **2023-04-17** | Partner Sarah Guo departed to launch her own AI-focused fund Conviction\n- **2023-08-25** | Led seed round for stealth AI infrastructure startup\n- **2024-01-12** | Hosted annual Greylock Techfair recruiting event for portfolio companies\n- **2024-06-03** | Published internal AI research report, shared selectively with LPs\n- **2024-11-19** | Co-invested with [Andreessen Horowitz](companies/a16z-3) in Series A for developer tools company\n- **2025-02-28** | Promoted two principals to partner, signaling generational transition",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/greylock-4",
-    "name": "Greylock",
-    "category": "vc",
-    "industry": "venture capital"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__gust-34.json b/eval/data/world-v1/companies__gust-34.json
deleted file mode 100644
index c1a349106..000000000
--- a/eval/data/world-v1/companies__gust-34.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "companies/gust-34",
-  "type": "company",
-  "title": "Gust",
-  "compiled_truth": "Gust is a data infrastructure startup founded in 2020 by [Steve Liu](people/steve-liu-34), who previously spent time at Snowflake and Databricks before striking out on his own. The company focuses on building real-time data pipelines that can handle massive throughput without the typical overhead of traditional ETL systems. Their core product lets engineering teams ingest, transform, and route streaming data with minimal configuration—think Kafka meets dbt but with a much simpler developer experience.\n\nThe founding story is pretty straightforward. Steve had grown frustrated with the complexity of existing data infrastructure tools while working on analytics pipelines at his previous roles. He saw an opportunity to build something cleaner, something that didn't require a dedicated platform team just to keep running. Gust was born out of that frustration, initially as a side project before Steve commited to it full-time.\n\nEarly traction came from mid-sized fintech companies who needed reliable streaming infrastructure but couldn't justify the headcount to manage Kafka clusters. Gust's managed offering hit a sweet spot—enterprise-grade reliability without the operational burden. By late 2021, the company had a handful of paying customers and was generating modest but growing revenue.\n\n[Sarah Lopez](people/sarah-lopez-84) led their seed round in early 2022, betting on Steve's technical chops and the growing demand for simplified data tooling. Sarah had been tracking the data infrastructure space for years and saw Gust as a potential breakout player. Her investment gave the company runway to expand the engineering team and accelerate product developement.\n\nToday Gust operates with a lean team of about 25 people, mostly engineers. They've been deliberate about not over-hiring, preferring to stay focused and capital-efficient. The company has expanded its product to include schema management, data quality monitoring, and connectors for most major data warehouses. Competition from bigger players like Confluent and newer startups remains intense, but Gust has carved out a loyal customer base that values simplicity over feature bloat.",
-  "timeline": "- **2020-03-15** | Steve Liu incorporates Gust and begins building the initial prototype\n- **2020-09-22** | First beta customer signs up—a small fintech startup in NYC\n- **2021-04-10** | Gust launches publicly with support for Postgres and Snowflake sinks\n- **2022-02-08** | Closes $4.2M seed round led by [Sarah Lopez](people/sarah-lopez-84)\n- **2022-07-19** | Hires first head of engineering from Stripe\n- **2023-01-30** | Launches schema registry feature after months of customer requests\n- **2023-11-14** | [Steve Liu](people/steve-liu-34) speaks at Data Council on simplifying streaming architectures\n- **2024-05-02** | Crosses 100 paying customers milestone\n- **2024-12-11** | Announces partnership with major cloud provider for native integration\n- **2025-08-20** | Begins work on Series A fundraising process",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/gust-34",
-    "name": "Gust",
-    "category": "startup",
-    "industry": "data infrastructure",
-    "founded_year": 2020,
-    "founders": [
-      "people/steve-liu-34"
-    ],
-    "investors": [
-      "people/sarah-lopez-84"
-    ],
-    "employees": [
-      "people/xavier-jackson-144"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__hatch-35.json b/eval/data/world-v1/companies__hatch-35.json
deleted file mode 100644
index 9223d3979..000000000
--- a/eval/data/world-v1/companies__hatch-35.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "companies/hatch-35",
-  "type": "company",
-  "title": "Hatch",
-  "compiled_truth": "Hatch is an edtech startup founded in 2019 by [Eric Miller](people/eric-miller-35), who saw an opportunity to reimagine how young professionals develop career skills outside traditional academic settings. The company operates in the increasingly crowded learn-to-earn space, but distinguishes itself through a cohort-based model that emphasizes peer accountability and real-world project work.\n\nThe platform connects early-career workers with mentors from established companies, facilitating structured 8-week programs in areas like product management, data analytics, and business development. Hatch takes a different aproach than most competitors—rather than selling courses to individuals, they partner directly with employers who want to upskill entry-level hires or create alternative talent pipelines. This B2B focus has given them more predictable revenue, though it's also meant slower user growth compared to consumer-facing platforms.\n\n[Steve Martinez](people/steve-martinez-192) joined as an advisor sometime in 2022, bringing his network in workforce development and helping Hatch refine their enterprise sales motion. His involvement signaled a shift toward targeting larger organizations rather than the SMB market they'd initially pursued. Martinez has been particularly helpful in opening doors at companies looking to diversify their hiring beyond traditional university recruiting.\n\nEric Miller remains the driving force behind product decisions. He's known for being hands-on with curriculum design, often personally reviewing program content and sitting in on mentor sessions. Some employees find this level of involvement micromanage-y, but others appreciate the attention to quality. The company has stayed relatively lean—around 35 employees as of late 2024—and Miller has been vocal about not raising more capital than necessary.\n\nHatch completed a Series A in early 2023, though they haven't disclosed the amount publicly. They're headquartered in Austin but operate fully remote, with mentors and participants spread across North America. Recent moves suggest they're exploring expansion into technical skills training, potentially competing more directly with bootcamps.",
-  "timeline": "- **2019-06-12** | Hatch incorporated in Delaware; [Eric Miller](people/eric-miller-35) begins building initial prototype\n- **2020-03-08** | Launched first pilot cohort with 24 participants across three employer partners\n- **2021-09-15** | Closed seed round of $2.4M led by Reach Capital\n- **2022-04-22** | [Steve Martinez](people/steve-martinez-192) formally joins advisory board\n- **2022-11-03** | Surpassed 2,000 program graduates; announced partnership with two Fortune 500 retailers\n- **2023-02-17** | Series A closed; terms undisclosed but reportedly in $8-12M range\n- **2023-08-29** | Launched data analytics track, first technical program offering\n- **2024-01-14** | Eric Miller spoke at ASU+GSV Summit on alternative credentialing\n- **2024-07-20** | Opened pilot in Canada with three Toronto-based employers\n- **2025-03-11** | Announced curriculum partnership with major cloud provider for technical upskilling",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/hatch-35",
-    "name": "Hatch",
-    "category": "startup",
-    "industry": "edtech",
-    "founded_year": 2019,
-    "founders": [
-      "people/eric-miller-35"
-    ],
-    "employees": [
-      "people/diana-brown-145"
-    ],
-    "advisors": [
-      "people/steve-martinez-192"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__helix-9.json b/eval/data/world-v1/companies__helix-9.json
deleted file mode 100644
index 394270bea..000000000
--- a/eval/data/world-v1/companies__helix-9.json
+++ /dev/null
@@ -1,26 +0,0 @@
-{
-  "slug": "companies/helix-9",
-  "type": "company",
-  "title": "Helix",
-  "compiled_truth": "Helix is an AI infrastructure startup founded in 2021 by [Rachel Garcia](people/rachel-garcia-9), a veteran systems engineer who previously led distributed computing teams at major cloud providers. The company focuses on building foundational tooling for deploying and managing large-scale machine learning workloads, with particular emphasis on GPU orchestration and model serving optimization.\n\nThe core product is a Kubernetes-native platform that abstracts away much of the complexity involved in running inference at scale. Helix's approach differs from competitors in that it prioritizes cost efficiency over raw performance—their scheduling algorithms are designed to maximize GPU utilization across heterogenous hardware, which appeals to companies running mixed fleets of older and newer accelerators. Early customers include several mid-size fintech firms and a handful of healthcare AI startups.\n\nRachel Garcia serves as CEO and has been the public face of the company since launch. She's known for her pragmatic approach to infrastructure problems and has spoken at several industry conferences about the \"unsexy\" challenges of ML ops. Under her leadership, Helix has grown to roughly 35 employees, mostly engineers with backgrounds in distributed systems and cloud infrastucture.\n\nThe advisory board includes [Xavier Patel](people/xavier-patel-183), who brings deep expertise in enterprise sales and go-to-market strategy, and [Bob Chen](people/bob-chen-185), a technical advisor with experience scaling infrastructure at hypergrowth companies. Both have been instrumental in shaping Helix's enterprise positioning.\n\nHelix raised a Series A in early 2023, though the company has been relatively quiet about specific metrics. Industry observers note that the AI infrastructure space has become increasingly crowded, but Helix's focus on cost optimization rather than cutting-edge performance gives it a distinct niche. The startup has been expanding its sales team and recently opened a small office in Austin to complement its San Francisco headquarters. Recent product updates have focused on observability features and tighter integrations with popular ML frameworks.",
-  "timeline": "- **2021-03-15** | Company incorporated by [Rachel Garcia](people/rachel-garcia-9) in Delaware\n- **2021-09-02** | Closed $4.2M seed round led by Gradient Ventures\n- **2022-01-18** | First production customer goes live on Helix platform\n- **2022-07-11** | [Xavier Patel](people/xavier-patel-183) joins as advisor to help with enterprise strategy\n- **2023-02-28** | Announced Series A funding, expanded engineering team to 25\n- **2023-08-14** | [Bob Chen](people/bob-chen-185) joins advisory board\n- **2024-01-22** | Launched Helix Observe, new monitoring and cost analytics product\n- **2024-06-09** | Rachel Garcia keynotes at MLOps World conference in Austin\n- **2024-11-03** | Opened Austin office, announced plans to double sales team\n- **2025-04-17** | Partnership announced with major cloud provider for marketplace listing",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/helix-9",
-    "name": "Helix",
-    "category": "startup",
-    "industry": "AI infrastructure",
-    "founded_year": 2021,
-    "founders": [
-      "people/rachel-garcia-9"
-    ],
-    "employees": [
-      "people/quinn-park-119"
-    ],
-    "advisors": [
-      "people/xavier-patel-183",
-      "people/bob-chen-185",
-      "people/victor-smith-193"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__helix-labs-59.json b/eval/data/world-v1/companies__helix-labs-59.json
deleted file mode 100644
index 1c2a844d5..000000000
--- a/eval/data/world-v1/companies__helix-labs-59.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "companies/helix-labs-59",
-  "type": "company",
-  "title": "Helix Labs",
-  "compiled_truth": "Helix Labs is a cybersecurity startup founded in 2020 by [Bob Jackson](people/bob-jackson-59), a former penetration tester who spent nearly a decade at major defense contractors before striking out on his own. The company focuses on automated threat detection for mid-market enterprises, a segment Jackson felt was underserved by existing solutions that either targeted Fortune 500 companies or were too basic for sophisticated threats.\n\nThe company's flagship product, HelixShield, uses behavioral analysis to identify anomalous network activity before breaches occur. Unlike traditional signature-based detection, their approach learns what 'normal' looks like for each client and flags deviations in real-time. Early customers have praised the low false-positive rate, though some have noted the onboarding process can be lengthy.\n\nHelix raised its seed round in late 2021 from angel investors including [Priya Taylor](people/priya-taylor-85) and [Julia Davis](people/julia-davis-86), both of whom have backgrounds in enterprise software. Priya in particular has been an active advisor, reportedly introducing the team to several key enterprise clients in the healthcare vertical. The company closed a Series A in 2023, though terms were not publicly disclosed.\n\nThe team has grown to around 45 employees, with engineering concentrated in Austin and a small sales presence in New York. Jackson remains CEO and is known for his hands-on technical involvement—he still reviews major architecture decisions and ocasionally jumps into customer calls when things get hairy. Former colleagues describe him as demanding but fair, with a tendency to work late nights that sometimes sets unrealistic expectations for the rest of the team.\n\nHelix Labs has been relatively quiet in terms of press, preferring to let customer referrals drive growth rather than splashy marketing campaigns. That said, there's been some chatter about a potential expansion into cloud security posture management, which would put them in direct competition with larger players. Whether they have the resources to fight on multiple fronts remaind to be seen.",
-  "timeline": "- **2020-03-15** | Helix Labs incorporated in Delaware by [Bob Jackson](people/bob-jackson-59)\n- **2020-09-22** | First prototype of HelixShield deployed internally for testing\n- **2021-06-10** | Closed seed round with participation from [Priya Taylor](people/priya-taylor-85) and [Julia Davis](people/julia-davis-86)\n- **2021-11-03** | Landed first paying customer, a regional hospital network in Texas\n- **2022-04-18** | Expanded engineering team to 20 people, opened Austin office\n- **2023-02-27** | Series A closed; valuation undisclosed but rumored around $40M\n- **2023-09-14** | HelixShield 2.0 launched with improved ML detection pipeline\n- **2024-05-06** | [Bob Jackson](people/bob-jackson-59) spoke at RSA Conference on behavioral threat detection\n- **2025-01-22** | Announced partnership with managed security provider NorthWatch\n- **2025-08-30** | Internal planning meetings hint at cloud security product expansion",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/helix-labs-59",
-    "name": "Helix Labs",
-    "category": "startup",
-    "industry": "cybersecurity",
-    "founded_year": 2020,
-    "founders": [
-      "people/bob-jackson-59"
-    ],
-    "investors": [
-      "people/priya-taylor-85",
-      "people/julia-davis-86"
-    ],
-    "employees": [
-      "people/sam-wilson-169"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__index-ventures-7.json b/eval/data/world-v1/companies__index-ventures-7.json
deleted file mode 100644
index 84e0531a7..000000000
--- a/eval/data/world-v1/companies__index-ventures-7.json
+++ /dev/null
@@ -1,14 +0,0 @@
-{
-  "slug": "companies/index-ventures-7",
-  "type": "company",
-  "title": "Index Ventures",
-  "compiled_truth": "Index Ventures is one of Europe's most storied venture capital firms, with a track record that spans three decades and includes some of the most consequential technology companies of the modern era. Founded in Geneva in 1996, the firm has grown to operate across offices in San Francisco, London, and Geneva, positioning itself as a truly transatlantic investor with deep roots on both sides of the pond.\n\nThe firm operates across multiple stages, from seed through growth, and has backed companies like Figma, Discord, Notion, Roblox, and Deliveroo. Index made early bets on European champions like Skype and King Digital, establishing its reputation for identifying category-defining companies before they hit mainstream radar. Their portfolio reflects a broad thesis covering enterprise software, fintech, consumer internet, and increasingly, AI-native applications.\n\nIndex is known for its partnership-driven model, where partners maintain significant autonomy in dealmaking while sharing economics equally. Notable partners include Danny Rimer, who led investments in Dropbox and Glossier, and Mike Volpi, a former Cisco executive who's become one of the most respected enterprise investors in the industry. The firm's approach tends to be founder-friendly, often taking board seats but avoiding the heavy-handed governance that characterizes some of their peers.\n\nRecent years have seen Index raising substantial funds—their 2021 vintage exceeded $3 billion across seed and growth vehicles. They've been particularly active in the AI infrastructure space, competing aggressively with firms like [Sequoia Capital](companies/sequoia-capital-12) for the hottest deals. Some partners have noted tension between maintaining their European identity while increasingly deploying capital into Silicon Valley's AI boom.\n\nThe firm has also made notable investments alongside [Andreessen Horowitz](companies/andreessen-horowitz-9) in several high-profile rounds, demonstrating their ability to co-invest with top-tier American firms while maintaining deal leadership. Index's LP base includes major endowments, sovereign wealth funds, and family offices who've stuck with the firm through multiple fund cycles.\n\nCriticism sometimes surfaces around their growth-stage valuations—some observers argue Index overpaid during the 2021 bubble. But their seed practice has remained disciplined, and their multi-stage model provides natural follow-on optionality that pure-play seed funds lack.",
-  "timeline": "- **2021-03-15** | Closed Index Ventures Growth VI at $2.3B, largest fund in firm history\n- **2021-09-22** | Led $150M Series C for AI startup alongside [Sequoia Capital](companies/sequoia-capital-12)\n- **2022-04-10** | Partner Martin Mignot promoted to lead European seed practice\n- **2022-11-08** | Portfolio company Figma announced $20B acquisition by Adobe (later terminated)\n- **2023-02-14** | Participated in Discord's down round, maintaining pro-rata\n- **2023-08-30** | Co-led infrastructure deal with [Andreessen Horowitz](companies/andreessen-horowitz-9) at $800M valuation\n- **2024-01-19** | Published annual European tech ecosystem report showing record unicorn creation\n- **2024-06-05** | Danny Rimer keynoted at Index's annual founder summit in London\n- **2025-02-28** | Announced new $1.8B early-stage fund focused on AI-native applications\n- **2025-09-12** | Opened small Tel Aviv office to expand Middle East dealflow",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/index-ventures-7",
-    "name": "Index Ventures",
-    "category": "vc",
-    "industry": "venture capital"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__initialized-11.json b/eval/data/world-v1/companies__initialized-11.json
deleted file mode 100644
index ca214016e..000000000
--- a/eval/data/world-v1/companies__initialized-11.json
+++ /dev/null
@@ -1,14 +0,0 @@
-{
-  "slug": "companies/initialized-11",
-  "type": "company",
-  "title": "Initialized Capital",
-  "compiled_truth": "Initialized Capital is a seed-stage venture capital firm that made a significant mark on Silicon Valley's early-stage investing landscape. Founded in 2011 by Alexis Ohanian and Garry Tan, the firm quickly established itself as a go-to partner for ambitious founders building transformative companies. Initialized became known for writing the first checks into startups that would go on to become household names.\n\nThe firm's portfolio included some remarkable successes. Coinbase, Instacart, Cruise Automation, and Flexport all received early backing from Initialized, demonstrating the partners' ability to identify breakout opportunities before they became obvious. The fund's investment thesis centered on backing technical founders with strong product instincts, often at the pre-seed or seed stage when most institutional investors wouldn't engage.\n\nGarry Tan served as managing partner and was the driving force behind much of the firm's deal flow and investment decisions. His background as a founder (he co-founded Posterous) and his time as a partner at Y Combinator gave him unique insight into what makes early-stage companies succeed. In 2022, Tan departed Initialized to take on the role of President and CEO at [Y Combinator](companies/y-combinator), leaving the firm at an inflection point.\n\nFollowing Tan's departure, the future of Initalized became somewhat uncertain. The firm had raised multiple funds over the years, with later vehicles exceeding $300 million in committed capital. Some partners continued to manage existing investments while the firm's active deployment slowed considerably.\n\nInitialized was part of a broader wave of seed-focused firms that emerged in the early 2010s, alongside peers like First Round Capital and [Floodgate](companies/floodgate). These micro-VCs helped fill a gap left by larger funds that had moved upstream to Series A and beyond. The firm's legacy lives on through its portfolio companies, many of wich continue to shape their respective industries. Alexis Ohanian has since focused his attention on other ventures, including Seven Seven Six, his newer investment vehicle.",
-  "timeline": "- **2011-06-15** | Initialized Capital founded by Alexis Ohanian and Garry Tan with a focus on seed-stage investments\n- **2017-03-22** | Closed Fund III at $225 million, marking significant growth from earlier vehicles\n- **2019-09-10** | Portfolio company Coinbase valuation exceeds $8 billion following private funding round\n- **2021-04-14** | Coinbase direct listing on NASDAQ delivers massive returns for early Initialized investment\n- **2022-01-18** | Garry Tan announced as incoming CEO of [Y Combinator](companies/y-combinator), signaling transition at Initialized\n- **2022-03-01** | Tan officially departs managing partner role to lead YC full-time\n- **2023-08-12** | Firm continues managing existing portfolio with reduced new investment activity\n- **2024-02-28** | Several Initialized portfolio companies announce down rounds amid market correction\n- **2025-05-14** | Legacy fund distributions continue as mature portfolio companies reach liquidity events",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/initialized-11",
-    "name": "Initialized",
-    "category": "vc",
-    "industry": "venture capital"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__iris-36.json b/eval/data/world-v1/companies__iris-36.json
deleted file mode 100644
index fb2765325..000000000
--- a/eval/data/world-v1/companies__iris-36.json
+++ /dev/null
@@ -1,27 +0,0 @@
-{
-  "slug": "companies/iris-36",
-  "type": "company",
-  "title": "Iris",
-  "compiled_truth": "Iris is a consumer social startup founded in 2024 by [Mia Park](people/mia-park-36), a first-time founder with a background in behavioral psychology and product design. The company is building what it describes as a \"mood-first\" social platform—users share emotional states and context rather than polished photos or status updates. The core thesis is that Gen Z craves authenticity but existing platforms still incentivize performance. Iris flips that by making vulnerability the default.\n\nThe app launched in closed beta in late 2024, initially targeting college campuses on the West Coast. Early traction was promising, with retention numbers that caught the attention of several angel investors. [Jack Davis](people/jack-davis-89) led a pre-seed round, drawn to Mia's unconventional approach and the product's sticky engagement loops. He's been hands-on, joining weekly product reviews and pushing the team to nail the onboarding flow before scaling.\n\nIris operates with a lean team of five, mostly engineers and one designer Mia poached from her previous gig at a larger social app. The company runs out of a cramped co-working space in San Francisco's Mission district. Culture is intense but collaborative—Mia sets aggressive ship cycles but also mandates \"disconnect Fridays\" to prevent burnout. There's a scrappy energy to the operation.\n\n[David Kim](people/david-kim-186) serves as an advisor, providing strategic guidence on growth tactics and helping Mia navigate the fundraising landscape. He's introduced her to several potential Series A leads, though the company isn't actively raising yet. The plan is to hit 100k MAU before pursuing a priced round.\n\nRecent product moves include a \"resonance\" feature that matches users with strangers experiencing similar emotional states. It's controversial internally—some worry about safety implications—but early data shows it drives significent engagement. Mia has publicly stated that Iris will never sell emotional data to advertisers, a stance that's resonated with privacy-conscious users but raises questions about eventual monetization.",
-  "timeline": "- **2024-01-15** | [Mia Park](people/mia-park-36) incorporates Iris and begins recruiting founding team\n- **2024-03-22** | Closed alpha launches with 200 users from Stanford and Berkeley\n- **2024-05-10** | [Jack Davis](people/jack-davis-89) commits to leading pre-seed round after demo day pitch\n- **2024-06-01** | Pre-seed closes at $1.2M, valuation undisclosed\n- **2024-08-14** | [David Kim](people/david-kim-186) joins as formal advisor\n- **2024-10-03** | Beta expands to 12 universities across California and Oregon\n- **2024-11-19** | \"Resonance\" feature ships, driving 40% increase in daily sessions\n- **2025-01-08** | Iris hits 25k monthly active users milestone\n- **2025-02-20** | Mia speaks at a consumer social meetup in SF about emotional-first design\n- **2025-04-12** | Company begins exploratory conversations with Series A investors",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/iris-36",
-    "name": "Iris",
-    "category": "startup",
-    "industry": "consumer social",
-    "founded_year": 2024,
-    "founders": [
-      "people/mia-park-36"
-    ],
-    "investors": [
-      "people/jack-davis-89"
-    ],
-    "employees": [
-      "people/david-anderson-146"
-    ],
-    "advisors": [
-      "people/david-kim-186"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__jolt-37.json b/eval/data/world-v1/companies__jolt-37.json
deleted file mode 100644
index cb5ca2061..000000000
--- a/eval/data/world-v1/companies__jolt-37.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "companies/jolt-37",
-  "type": "company",
-  "title": "Jolt - AI Applications Startup",
-  "compiled_truth": "Jolt is an early-stage startup founded in 2025 by [Chris Williams](people/chris-williams-37), operating in the AI applications space. The company emerged during a particularly competitive period for AI ventures, yet managed to secure backing from notable angel investors including [Tina Hernandez](people/tina-hernandez-97) and [Chris Miller](people/chris-miller-101).\n\nThe company focuses on building AI-powered productivity tools aimed at small and medium businesses. Their flagship product, still in development, promises to automate routine administrative tasks using a combination of large language models and custom workflow engines. Chris Williams has described the vision as \"AI that actually fits into how people already work, not the other way around.\"\n\nJolt operates with a lean team, currently around 8 people, mostly engineers with backgrounds in ML infrastructure and frontend development. The company maintains offices in Austin, though most of the team works remotley. Williams has been vocal about keeping the team small until they achieve stronger product-market fit, a philosophy he picked up from his previous startup experience.\n\nFunding details remain somewhat private, but sources suggest the initial round was in the $2-3M range. [Chris Miller](people/chris-miller-101) reportedly led the round after meeting Williams at a conference in late 2024. The investment thesis centered on Williams' track record and the team's technical depth rather than any revolutionary technology moat.\n\nThe startup has been relatively quiet publicly, preferring to focus on building rather than marketing. A private beta launched in Q1 2025 with around 50 companies participating. Early feedback has been mixed but promising—users appreciate the simplicity but want more integrations. The team is currently heads-down on expanding connector support for popular tools like Slack, Notion, and various CRMs.\n\nCompetition in the AI productivity space is fierce, with both well-funded startups and big tech players vying for attention. Jolt's bet is that their focus on SMBs and ease of deployment will carve out a defensible niche. Whether that pans out remains to be seen.",
-  "timeline": "- **2024-11-15** | Chris Williams meets [Chris Miller](people/chris-miller-101) at AI Summit Austin, initial discussions about Jolt concept\n- **2025-01-08** | Jolt officially incorporated in Delaware\n- **2025-01-22** | Seed round closes with participation from [Tina Hernandez](people/tina-hernandez-97) and Chris Miller\n- **2025-02-10** | First two engineers hired, both former colleagues of [Chris Williams](people/chris-williams-37)\n- **2025-03-05** | Internal alpha of core product completed\n- **2025-04-12** | Private beta launches with 50 SMB partners\n- **2025-05-20** | Team expands to 8 people, adds first dedicated product manager\n- **2025-06-18** | Partnership discussions begin with major CRM vendor\n- **2025-07-02** | Beta feedback review leads to pivot toward deeper integrations focus",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/jolt-37",
-    "name": "Jolt",
-    "category": "startup",
-    "industry": "AI applications",
-    "founded_year": 2025,
-    "founders": [
-      "people/chris-williams-37"
-    ],
-    "investors": [
-      "people/tina-hernandez-97",
-      "people/chris-miller-101"
-    ],
-    "employees": [
-      "people/xavier-johnson-147"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__keel-38.json b/eval/data/world-v1/companies__keel-38.json
deleted file mode 100644
index ef70c26dc..000000000
--- a/eval/data/world-v1/companies__keel-38.json
+++ /dev/null
@@ -1,27 +0,0 @@
-{
-  "slug": "companies/keel-38",
-  "type": "company",
-  "title": "Keel",
-  "compiled_truth": "Keel is a crypto startup founded in early 2025 by [Steve Williams](people/steve-williams-38), a serial entrepreneur with a background in decentralized finance protocols. The company operates in the digital asset infrastructure space, focusing on building institutional-grade custody and settlement solutions for blockchain networks. Despite being a newcomer to an already crowded market, Keel has positioned itself as a lean alternative to legacy crypto custodians, emphasizing speed and regulatory compliance from day one.\n\nThe founding thesis behind Keel centers on the belief that traditional crypto custody providers have become bloated and slow to adapt to emerging Layer 2 ecosystems. Steve Williams has been vocal about this gap, arguing that institutions need nimble partners who understand the nuances of rollups, bridges, and cross-chain liquidity. The company's initial product focuses on Ethereum L2 settlement, with plans to expand into Bitcoin sidechains by late 2025.\n\nKeel raised a pre-seed round in Q1 2025, with [Carol Jackson](people/carol-jackson-81) serving as the lead investor. Jackson, known for her contrarian bets in fintech infrastructure, apparently saw potential in Williams' vision despite the bear market sentiment still lingering from 2024. The round was modest—reportedly under $3 million—but gave the team runway to build out their core platform and hire a small enginering team.\n\nAdvisory support comes from [Linda Taylor](people/linda-taylor-178), who brings regulatory expertise to the table. Taylor's involvement signals that Keel is serious about compliance, a differentiator in an industry still grappling with enforcement actions. Her guidance has reportedly shaped the company's approach to KYC/AML integration and its conversations with potential banking partners.\n\nThe team remains small, operating out of a co-working space in Austin. Williams has kept headcount intentionally low, preferring to ship fast with a tight-knit group rather than scale prematurely. Early users include a handful of crypto-native hedge funds testing the settlement infrastucture in sandbox environments. Keel's public launch is expected sometime in Q3 2025.",
-  "timeline": "- **2024-11-15** | Steve Williams begins exploratory conversations with early backers about a new custody venture\n- **2025-01-08** | Keel officially incorporated in Delaware; [Steve Williams](people/steve-williams-38) named CEO\n- **2025-01-22** | [Carol Jackson](people/carol-jackson-81) commits to leading the pre-seed round\n- **2025-02-10** | Pre-seed funding closes at $2.8M; team begins hiring engineers\n- **2025-02-28** | [Linda Taylor](people/linda-taylor-178) joins as regulatory advisor\n- **2025-03-15** | First internal demo of L2 settlement prototype completed\n- **2025-04-02** | Keel signs NDA with two crypto hedge funds for pilot testing\n- **2025-05-19** | Williams speaks at ETH Denver satellite event on institutional DeFi infrastructure\n- **2025-06-07** | Sandbox testing begins with select institutional partners",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/keel-38",
-    "name": "Keel",
-    "category": "startup",
-    "industry": "crypto",
-    "founded_year": 2025,
-    "founders": [
-      "people/steve-williams-38"
-    ],
-    "investors": [
-      "people/carol-jackson-81"
-    ],
-    "employees": [
-      "people/zoe-nakamura-148"
-    ],
-    "advisors": [
-      "people/linda-taylor-178"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__khosla-ventures-8.json b/eval/data/world-v1/companies__khosla-ventures-8.json
deleted file mode 100644
index 0da21ee85..000000000
--- a/eval/data/world-v1/companies__khosla-ventures-8.json
+++ /dev/null
@@ -1,14 +0,0 @@
-{
-  "slug": "companies/khosla-ventures-8",
-  "type": "company",
-  "title": "Khosla Ventures",
-  "compiled_truth": "Khosla Ventures is a prominent Silicon Valley venture capital firm founded in 2004 by Vinod Khosla, a co-founder of Sun Microsystems. The firm has established itself as one of the most influential investors in technology and cleantech, with a particular focus on companies that can have transformative impact across industries. Headquartered in Menlo Park, California, Khosla operates with a distinctive philosophy that embraces high-risk, high-reward bets on unproven technologies.\n\nThe firm manages multiple funds totaling billions in assets under managment, including seed funds for earlier-stage investments and larger growth funds for follow-on financing. Khosla Ventures has backed some notable successes including Square, DoorDash, and Instacart. More recently, the firm has been aggressively investing in artificial intelligence infrastructure and applications, recognizing the generational shift hapening in enterprise software.\n\nVinod Khosla himself remains deeply involved in investment decisions and is known for his contrarian views and willingness to fund moonshot ideas. The firm's team includes partners with deep technical backgrounds, which allows them to evaluate complex technologies that other VCs might shy away from. They've developed a reputation for being founder-friendly while also providing substantial operational support.\n\nKhosla Ventures has been particularly active in climate tech, betting big on carbon capture, alternative proteins, and next-generation energy storage. This aligns with Vinod's long-standing interest in technologies that address major societal challenges. The firm often co-invests alongside other major venture players like [Andreessen Horowitz](companies/a16z) on larger rounds, though they're equally comfortable leading deals solo.\n\nTheir investment approach tends to be thesis-driven rather than opportunistic. Partners develop deep conviction around specific technology shifts and then actively seek out founders building in those areas. This has led to early positions in categories before they become crowded. The firm maintains close relationships with the Stanford ecosystem and frequently backs technical founders straight out of PhD programs. Recent portfolio companies have explored everything from quantum computing to synthetic biology, reflecting Khosla's continued appetite for frontier tech bets.",
-  "timeline": "- **2021-03-15** | Khosla Ventures closed Fund VII at $1.4 billion, oversubscribed due to strong LP demand\n- **2021-09-22** | Led $50M Series B in carbon removal startup, signaling renewed climate focus\n- **2022-04-08** | Vinod Khosla keynoted Stanford entrepreneurship conference on AI's transformative potential\n- **2022-11-30** | Announced strategic partnership with [Andreessen Horowitz](companies/a16z) for joint investment in AI infrastructure deals\n- **2023-06-14** | Portfolio company Impossible Foods explored IPO options with firm's guidance\n- **2023-12-01** | Khosla published annual predictions letter, forecasting major disruption in healthcare from AI diagnostics\n- **2024-05-19** | Promoted two new general partners from within, expanding investment team to twelve\n- **2024-09-03** | Led $120M growth round for enterprise AI startup at $900M valuation\n- **2025-02-28** | Filed for Fund VIII targeting $2.1 billion across seed and growth vehicles\n- **2025-08-11** | Hosted annual LP summit in Palo Alto featuring portfolio company demos",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/khosla-ventures-8",
-    "name": "Khosla Ventures",
-    "category": "vc",
-    "industry": "venture capital"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__kindle-20.json b/eval/data/world-v1/companies__kindle-20.json
deleted file mode 100644
index 9eaa33f8b..000000000
--- a/eval/data/world-v1/companies__kindle-20.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "companies/kindle-20",
-  "type": "company",
-  "title": "Kindle - Climate Tech Startup",
-  "compiled_truth": "Kindle is a climate tech startup founded in 2023 by [Vera Singh](people/vera-singh-20), focused on developing next-generation carbon capture solutions for industrial emitters. The company emerged from Singh's doctoral research at MIT, where she pioneered novel membrane technologies that significantly reduce the energy costs of direct air capture.\n\nThe startup operates out of Oakland, California, with a small but growing team of around 15 engineers and scientists. Kindle's core product is a modular carbon capture unit designed for mid-sized manufacturing facilities—a market segment that's been largely overlooked by bigger players chasing utility-scale deployments. Their approach prioritizes affordability and ease of installation over raw capture volume, betting that widespread adoption matters more than individual unit performance.\n\nKindle has attracted notable advisors including [Tina Moore](people/tina-moore-191), who brings decades of experience scaling hardware startups. Moore's involvement has been particularly valuable in helping the company navigate supply chain challenges and establish early manufacturing partnerships. The advisory relationship reportedly began after a chance meeting at a climate conference in late 2023.\n\nThe company closed a seed round in early 2024, though exact figures haven't been publicly disclosed. Industry sources suggest somewhere in the $4-6M range, with participation from several climate-focused VCs and a strategic investment from a major cement manufacturer. Vera has been quoted saying the cement partnership represents exactly the kind of industrial collaboration Kindle needs to prove out thier technology at scale.\n\nRecent activity suggests Kindle is preparing for pilot deployments at two manufacturing sites in the midwest, with plans to gather operational data through 2025. The team has been hiring aggressivley for field engineering roles, a sign that real-world testing is imminent. Competition in the carbon capture space remains fierce, but Kindle's focus on the underserved mid-market could give them a meaningful niche if execution goes well.",
-  "timeline": "- **2023-03-15** | Kindle incorporated in Delaware by founder [Vera Singh](people/vera-singh-20)\n- **2023-06-22** | First prototype membrane unit achieves 40% efficiency improvement over baseline\n- **2023-11-08** | [Tina Moore](people/tina-moore-191) joins as lead advisor following Climate Forward conference\n- **2024-01-30** | Seed funding round closed with climate-focused VC syndicate\n- **2024-04-12** | Strategic partnership announced with Midwest cement manufacturer\n- **2024-07-19** | Team expands to 15 employees, opens Oakland R&D facility\n- **2024-10-03** | Vera Singh presents at TechCrunch Disrupt climate track\n- **2025-02-14** | Pilot deployment begins at first manufacturing partner site\n- **2025-05-20** | Second pilot location confirmed in Ohio",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/kindle-20",
-    "name": "Kindle",
-    "category": "startup",
-    "industry": "climate tech",
-    "founded_year": 2023,
-    "founders": [
-      "people/vera-singh-20"
-    ],
-    "employees": [
-      "people/julia-jones-130"
-    ],
-    "advisors": [
-      "people/tina-moore-191"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__kleiner-perkins-14.json b/eval/data/world-v1/companies__kleiner-perkins-14.json
deleted file mode 100644
index e437cfb3f..000000000
--- a/eval/data/world-v1/companies__kleiner-perkins-14.json
+++ /dev/null
@@ -1,14 +0,0 @@
-{
-  "slug": "companies/kleiner-perkins-14",
-  "type": "company",
-  "title": "Kleiner Perkins",
-  "compiled_truth": "Kleiner Perkins is one of the most storied venture capital firms in Silicon Valley, with a legacy stretching back to 1972. Founded by Eugene Kleiner and Tom Perkins, the firm helped shape the modern tech landscape through early bets on companies like Amazon, Google, and Genentech. Today, KP continues to operate as a top-tier growth and early-stage investor, though its position has evolved considerably from its peak influence in the 1990s and 2000s.\n\nThe firm operates primarily out of Menlo Park, California, maintaining a relatively focused team compared to mega-funds like Andreessen Horowitz or Sequoia. Kleiner Perkins has historically been organized around sector-specific practices, including digital health, fintech, enterprise, and consumer technology. Recent years have seen the firm double down on AI and machine learning opportunities, recognizing the transformative potential of foundation models and applied AI startups.\n\nNotable current partners include Mamoon Hamid, who joined from Social Capital, and Bucky Moore, known for his work in enterprise software. The firm has maintained relationships with iconic founders and frequently co-invests alongside other major players in the ecosystem. Their portfolio includes breakout successes like Figma, Rippling, and several emerging AI-native companies that are reshaping enterprise workflows.\n\nKleiner's approach to venture has shifted somewhat over the past decade. After struggling with its green tech investments in the early 2010s, the firm refocused on software and healthcare, areas where it had demonstrated repeateable success. The cleantech experiment, while producing some winners, largely taught KP hard lessons about capital intensity and market timing. They've since been more disciplined about sector allocation.\n\nThe firm typically writes checks ranging from $1M to $50M depending on stage, though they've participated in larger rounds for high-conviction bets. KP maintains a builder-friendly reputation, often providing operational support through its platform team and network of advisors. They host regular founder dinners and have been known to facilitate introductions across their portfolio companies.\n\nAs of 2024, Kleiner Perkins manages several billion dollars across multiple funds, continuing to attract institutional LPs despite increased competition in the venture landscape. The firm remains a sought-after partner for founders seeking both capital and credibility, though they face stiff competiton from newer entrants with aggressive deployment strategies.",
-  "timeline": "- **2021-03-15** | Kleiner Perkins closed Fund XX at $1.8B, marking a return to larger fund sizes after years of more modest raises.\n- **2021-09-22** | Led Series B for an AI-native workflow automation startup, signaling renewed focus on enterprise machine learning applications.\n- **2022-04-08** | Partner Bucky Moore spoke at a founders summit on the future of vertical SaaS and embedded fintech.\n- **2022-11-30** | KP participated in Figma's final private round before the Adobe acquisition announcement.\n- **2023-06-14** | Announced new partner hire from Stripe, expanding fintech and payments expertise within the firm.\n- **2023-10-02** | Hosted annual CEO Summit in Napa Valley, bringing together portfolio founders for networking and strategy sessions.\n- **2024-02-19** | Led $40M Series A for a foundation model fine-tuning platform focused on healthcare applications.\n- **2024-08-07** | Kleiner Perkins published research report on AI agent adoption trends across enterprise customers.\n- **2025-01-23** | Participated in growth round for Rippling, continuing long-standing relationship with Parker Conrad.\n- **2025-05-11** | Mamoon Hamid joined board of a stealth climate software startup, marking selective return to climate-adjacent investments.",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/kleiner-perkins-14",
-    "name": "Kleiner Perkins",
-    "category": "vc",
-    "industry": "venture capital"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__lattice-39.json b/eval/data/world-v1/companies__lattice-39.json
deleted file mode 100644
index 1173d26d9..000000000
--- a/eval/data/world-v1/companies__lattice-39.json
+++ /dev/null
@@ -1,27 +0,0 @@
-{
-  "slug": "companies/lattice-39",
-  "type": "company",
-  "title": "Lattice - Enterprise SaaS Startup",
-  "compiled_truth": "Lattice is an enterprise SaaS startup founded in 2022 by [Quinn Miller](people/quinn-miller-39), a repeat founder with a background in developer tools and infrastructure software. The company focuses on building next-generation workflow automation platfroms for mid-market and enterprise customers, specifically targeting operations teams who struggle with fragmented tooling across their organizations.\n\nThe company emerged from Quinn's frustration with existing solutions that either served small teams or required massive implementation budgets. Lattice positions itself in the middle ground—powerful enough for complex enterprise needs, but accessible enough that a single ops manager can get started without a consulting engagement. Their core product offers visual workflow builders, deep integrations with popular SaaS tools, and an AI-assisted configuration layer that helps users identify automation opportunities.\n\nEarly backing came from [Vera Gonzalez](people/vera-gonzalez-103), who led a seed round in late 2022. Vera had previously invested in several successful enterprise software companies and saw Lattice as addressing a genuine gap in the market. The company has since grown to approximately 25 employees, with engineering and product teams based primarily in San Francisco.\n\nOn the advisory side, Lattice brought on [Steve Martinez](people/steve-martinez-192) to help navigate enterprise sales cycles and GTM strategy. Steve's experience scaling sales organizations has proven valuable as Lattice transitions from founder-led sales to building out a dedicated revenue team. His connections in the Fortune 500 have also opened doors for pilot conversations that would otherwise take months to secure.\n\nLattice has been relatively quiet publicly, preferring to focus on product development and early customer success over PR. However, industry insiders note that the company has secured several notable design partners in the fintech and healthcare sectors. Their approach emphasizes landing with a single team and expanding organically—a strategy that keeps churn low but requires patience on revenue growth. The company is currently preparing for a Series A raise expected sometime in mid-2025.",
-  "timeline": "- **2022-03-15** | [Quinn Miller](people/quinn-miller-39) incorporates Lattice and begins initial product development\n- **2022-09-22** | Closes $3.2M seed round led by [Vera Gonzalez](people/vera-gonzalez-103)\n- **2022-12-01** | First design partner signed—a mid-sized fintech processing loan applications\n- **2023-04-18** | [Steve Martinez](people/steve-martinez-192) joins as formal advisor to help build sales playbook\n- **2023-08-30** | Launches private beta with 12 companies participating\n- **2024-01-15** | Reaches $500K ARR milestone, transitions to general availability\n- **2024-06-12** | Expands integration library to cover 80+ enterprise tools\n- **2024-11-03** | Hires first dedicated VP of Sales, growing team to 25 employees\n- **2025-02-20** | Begins Series A fundraising conversations with top-tier VCs",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/lattice-39",
-    "name": "Lattice",
-    "category": "startup",
-    "industry": "enterprise SaaS",
-    "founded_year": 2022,
-    "founders": [
-      "people/quinn-miller-39"
-    ],
-    "investors": [
-      "people/vera-gonzalez-103"
-    ],
-    "employees": [
-      "people/owen-patel-149"
-    ],
-    "advisors": [
-      "people/steve-martinez-192"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__lightspeed-6.json b/eval/data/world-v1/companies__lightspeed-6.json
deleted file mode 100644
index 2a6bf90a1..000000000
--- a/eval/data/world-v1/companies__lightspeed-6.json
+++ /dev/null
@@ -1,14 +0,0 @@
-{
-  "slug": "companies/lightspeed-6",
-  "type": "company",
-  "title": "Lightspeed Venture Partners",
-  "compiled_truth": "Lightspeed Venture Partners is a global venture capital firm with a storied history dating back to 2000. The firm has established itself as one of the most influential players in early and growth-stage investing, with a particular strength in enterprise software, consumer internet, and fintech. Headquartered in Menlo Park, California, Lightspeed operates across multiple geographies including offices in India, China, Israel, and Europe.\n\nThe firm manages over $25 billion in committed capital across various funds and has backed some of the most consequential technology companies of the past two decades. Notable investments include Snap, Affirm, Mulesoft, and Rubrik. Lightspeed tends to take a hands-on approach with portfolio companies, often providing operational support and leveraging their extensive network to help founders scale.\n\nIn recent years, Lightspeed has been particularly agressive in the AI and machine learning space, deploying significant capital into foundational model companies and AI-native applications. The firm closed a $7.1 billion fund in 2022, one of the largest in its history, signaling continued confidence from LPs despite broader market uncertainty. Partners like Ravi Mhatre and Arif Janmohamed have been instrumental in shaping the firm's enterprise investing thesis.\n\nLightspeed has developed relationships with other major firms in the ecosystem, occasionally co-investing alongside [Andreessen Horowitz](companies/a16z) on competitive deals. The firm is known for moving quickly on conviction and has a reputation for being founder-friendly, though they maintain rigourous diligence processes. Their global footprint allows them to spot trends early—the India team, for instance, was early to companies like Oyo and Byju's before those markets became crowded.\n\nThe firm also runs Lightspeed Faction, a growth-stage vehicle that targets later rounds. This multi-stage capability has become increasingly important as companies stay private longer. They've competed for deals with firms like [Sequoia Capital](companies/sequoia) across multiple stages, sometimes winning on speed and sometimes on terms. Lightspeed remains a top-tier firm that consistently ranks among the most active investors globally.",
-  "timeline": "- **2021-03-15** | Lightspeed leads $150M Series C for enterprise AI startup, marking increased focus on machine learning infrastructure\n- **2021-09-22** | Announced expansion of Israel office with three new partner hires\n- **2022-04-10** | Closed $7.1 billion across early and growth funds, largest raise in firm history\n- **2022-11-08** | Co-invested alongside [Andreessen Horowitz](companies/a16z) in developer tools company seed round\n- **2023-02-14** | Published annual report showing 47 new investments across global portfolio in 2022\n- **2023-07-19** | Partner Mercedes Bent promoted to lead consumer investing practice\n- **2024-01-30** | Lightspeed Faction leads $200M growth round for cybersecurity unicorn\n- **2024-06-12** | Competed with [Sequoia Capital](companies/sequoia) for Series B deal in logistics automation space\n- **2025-02-28** | Opened new office in London to expand European coverage\n- **2025-09-05** | Announced $500M opportunity fund focused exclusively on AI applications",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/lightspeed-6",
-    "name": "Lightspeed",
-    "category": "vc",
-    "industry": "venture capital"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__lucid-21.json b/eval/data/world-v1/companies__lucid-21.json
deleted file mode 100644
index 63164e653..000000000
--- a/eval/data/world-v1/companies__lucid-21.json
+++ /dev/null
@@ -1,28 +0,0 @@
-{
-  "slug": "companies/lucid-21",
-  "type": "company",
-  "title": "Lucid",
-  "compiled_truth": "Lucid is a climate tech startup founded in 2020 by [Eric Lee](people/eric-lee-21), focused on developing next-generation carbon capture monitoring systems. The company emerged from Eric's frustration with the lack of real-time verification tools in the voluntary carbon markets—a gap he identified while working on sustainability initiatives at his previous role.\n\nThe core product is a hardware-software platform that provides continous monitoring of carbon sequestration projects, particularly direct air capture facilities and reforestation efforts. Lucid's sensors collect granular data on CO2 flux, which feeds into their analytics dashboard used by project developers, carbon credit buyers, and third-party verifiers. The pitch is simple: if you're buying carbon credits, you should know they're actually removing carbon.\n\nIn 2022, Lucid raised a seed round led by [Fiona Moore](people/fiona-moore-88), with participation from [Ian Anderson](people/ian-anderson-105). The round valued the company at roughly $18M and gave them runway to expand their pilot programs across North America. Fiona joined the board and has been instrumental in connecting Lucid to her network of institutional investors interested in climate infrastructure.\n\nThe company operates lean—around 25 employees as of late 2024, split between hardware engineering in Oakland and a software team that's mostly remote. [Vera Rodriguez](people/vera-rodriguez-171) serves as an advisor, bringing her expertise in carbon markets and regulatory frameworks. Her guidance has been particularly valuable as Lucid navigates the evolving landscape of carbon credit certification standards.\n\nLucid has faced some headwinds. The voluntary carbon market contracted in 2023 amid scrutiny over credit quality, which ironically validated Lucid's core thesis but also slowed sales cycles. Several potential enterprise deals got pushed as companies reassesed their offset strategies. Still, the team sees this as a temporary correction that ultimately benefits players focused on verification and transparency.\n\nRecent moves include a partnership with a major reforestation nonprofit to pilot their monitoring tech across 50,000 hectares in the Pacific Northwest. Eric has been increasingly visible at climate conferences, positioning Lucid as the \"trust layer\" for carbon markets.",
-  "timeline": "- **2020-06-15** | Lucid incorporated by [Eric Lee](people/eric-lee-21) in Delaware, initial focus on carbon monitoring R&D\n- **2021-03-22** | First prototype sensor deployed at a test site in Nevada desert\n- **2021-11-08** | Accepted into climate tech accelerator program, relocated operations to Oakland\n- **2022-04-30** | Closed $4.2M seed round led by [Fiona Moore](people/fiona-moore-88)\n- **2022-09-14** | Hired VP of Engineering from Planet Labs to scale hardware team\n- **2023-02-17** | [Vera Rodriguez](people/vera-rodriguez-171) formally joins as strategic advisor\n- **2023-08-05** | Eric presents at Climate Week NYC on verification standards\n- **2024-01-20** | Announced partnership with ForestWatch nonprofit for Pacific Northwest pilot\n- **2024-07-11** | Reached 15 active deployment sites across US and Canada\n- **2025-03-03** | Began Series A conversations, targeting $15-20M raise",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/lucid-21",
-    "name": "Lucid",
-    "category": "startup",
-    "industry": "climate tech",
-    "founded_year": 2020,
-    "founders": [
-      "people/eric-lee-21"
-    ],
-    "investors": [
-      "people/fiona-moore-88",
-      "people/ian-anderson-105"
-    ],
-    "employees": [
-      "people/ian-nakamura-131"
-    ],
-    "advisors": [
-      "people/vera-rodriguez-171"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__lumen-12.json b/eval/data/world-v1/companies__lumen-12.json
deleted file mode 100644
index d07741854..000000000
--- a/eval/data/world-v1/companies__lumen-12.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "companies/lumen-12",
-  "type": "company",
-  "title": "Lumen - Biotech Startup",
-  "compiled_truth": "Lumen is a biotech startup founded in 2018 by [Henry Johnson](people/henry-johnson-12), focused on developing novel diagnostic tools for early-stage cancer detection. The company operates out of Cambridge, Massachusetts, positioning itself within one of the most concentrated biotech ecosystems in the world. Their core technology leverages proprietary biomarker identification methods combined with machine learning to detect malignancies from standard blood draws—sometimes called liquid biopsy approaches.\n\nThe founding story traces back to Johnson's graduate research at MIT, where he first identified a unique protein signature associated with pancreatic cancer. Rather than pursue a traditional academic path, he spun out the research into what would become Lumen. Early days were scrappy. The company ran lean for nearly two years before securing meaningful outside investment.\n\nLumen's investor base includes [Kate Lopez](people/kate-lopez-99), who led their seed round in late 2020, and [Sarah Wang](people/sarah-wang-104), who joined during the Series A. Both have been activley involved in shaping company strategy, with Lopez taking a board observer seat and Wang providing introductions to pharmaceutical partners. The relationship with these backers has been described as collaborative rather than hands-off—monthly check-ins, strategic planning sessions, the works.\n\nOn the product side, Lumen has made steady progress. Their flagship diagnostic, LumenScreen, completed initial clinical validation in 2023 and is currently pursuing FDA breakthrough device designation. The team has grown to around 45 employees, split between R&D and clinical operations. They've also inked a partnership with a major regional hospital network for pilot testing, though terms weren't disclosed publically.\n\nHenry Johnson remains CEO and is known for a somewhat reserved public presence—he rarely speaks at conferences and prefers to let data do the talking. Internally, employees describe the culture as intense but mission-driven. Turnover has been relatively low for a company at this stage.\n\nLumen faces stiff competition from larger players in the liquid biopsy space, including Grail and Guardant Health. But the company's narrow focus on specific cancer types may prove advantageous for regulatory approval and clinical adoption. The next 18 months will be critical as they push toward commercialization.",
-  "timeline": "- **2018-03-15** | Lumen incorporated in Delaware by [Henry Johnson](people/henry-johnson-12)\n- **2018-09-22** | First lab space secured in Cambridge, initial team of 3 hired\n- **2020-11-08** | Seed round closed with [Kate Lopez](people/kate-lopez-99) leading at $2.4M\n- **2021-06-30** | Biomarker panel v1 validated in preclinical studies\n- **2022-04-12** | Series A announced, $18M raised with participation from [Sarah Wang](people/sarah-wang-104)\n- **2023-01-19** | LumenScreen enters clinical validation trials across 4 sites\n- **2023-08-07** | Partnership announced with Northeast Regional Health System for pilot deployment\n- **2024-02-28** | FDA breakthrough device designation application submitted\n- **2024-11-15** | Team expands to 45 full-time employees\n- **2025-03-22** | Preliminary data from clinical trials presented at AACR annual meeting",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/lumen-12",
-    "name": "Lumen",
-    "category": "startup",
-    "industry": "biotech",
-    "founded_year": 2018,
-    "founders": [
-      "people/henry-johnson-12"
-    ],
-    "investors": [
-      "people/kate-lopez-99",
-      "people/sarah-wang-104"
-    ],
-    "employees": [
-      "people/grace-miller-122"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__mantle-16.json b/eval/data/world-v1/companies__mantle-16.json
deleted file mode 100644
index fa0a88190..000000000
--- a/eval/data/world-v1/companies__mantle-16.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "companies/mantle-16",
-  "type": "company",
-  "title": "Mantle",
-  "compiled_truth": "Mantle is a consumer social startup founded in 2024 by [Ulrich Wang](people/ulrich-wang-16), an entrepreneur with a background in community-driven products. The company is building what they describe as a \"social layer for real-world experiences\" — essentially trying to bridge the gap between digital social graphs and physical gatherings. Early product demos have shown features around spontaneous meetups, location-based discovery, and ephemeral group chats tied to specific venues or events.\n\nThe founding team is lean, with Ulrich handling most of the product vision and early engineering. He's been advised by [Julia Wilson](people/julia-wilson-194), who brings experience from previous consumer social ventures and has been instrumental in shaping Mantle's go-to-market thinking. Julia's involvement suggests the company is serious about avoiding the common pitfalls of consumer social — namely, building features nobody asked for and failing to find organic growth loops.\n\nMantle's thesis is that existing social apps have become too performative, too oriented around content creation rather than genuine connection. The team believes there's an underserved segment of users who want lower-friction ways to coordinate IRL hangs without the pressure of posting or maintaining a public persona. It's a crowded space, but Wang argues that most competitors have gotten the incentive structures wrong — focusing on creator monetization when they should be focusing on social utility.\n\nThe company hasn't announced any funding publicly, though sources suggest they've raised a small pre-seed round from angels in the consumer space. Headcount remains under five as of late 2024. Mantle is currently testing with a closed beta group, primarly college students in the Bay Area and a few cities on the East Coast.\n\nWhether Mantle can break through remains to be seen. Consumer social is notoriously difficult — network effects cut both ways, and user attention is finite. But with Ulrich's obsessive focus on user experience and Julia Wilson's strategic guidance, the company has a shot at carving out a niche. Early retention numbers are reportedly encouraging, though the team is tight-lipped about specifics.",
-  "timeline": "- **2024-01-18** | Ulrich Wang incorporates Mantle as a Delaware C-corp, begins solo development on MVP.\n- **2024-03-02** | [Julia Wilson](people/julia-wilson-194) joins as an advisor after intro from a mutual investor.\n- **2024-04-15** | Mantle closes a small pre-seed round; terms undisclosed.\n- **2024-06-10** | First internal alpha launched to ~50 testers across three college campuses.\n- **2024-08-22** | Company hires first full-time engineer, a former classmate of [Ulrich Wang](people/ulrich-wang-16).\n- **2024-09-30** | Closed beta expands to 500 users; early retention data looks promising.\n- **2024-11-12** | Mantle presents at a small consumer social showcase in SF, generates some buzz.\n- **2025-01-08** | Team begins exploring partnerships with event venues for location-based features.\n- **2025-03-20** | Beta user count crosses 2,000; team considering seed raise timing.",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/mantle-16",
-    "name": "Mantle",
-    "category": "startup",
-    "industry": "consumer social",
-    "founded_year": 2024,
-    "founders": [
-      "people/ulrich-wang-16"
-    ],
-    "employees": [
-      "people/noah-lopez-126"
-    ],
-    "advisors": [
-      "people/julia-wilson-194"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__meridian-40.json b/eval/data/world-v1/companies__meridian-40.json
deleted file mode 100644
index f8cf576ee..000000000
--- a/eval/data/world-v1/companies__meridian-40.json
+++ /dev/null
@@ -1,29 +0,0 @@
-{
-  "slug": "companies/meridian-40",
-  "type": "company",
-  "title": "Meridian",
-  "compiled_truth": "Meridian is a developer tools startup founded in 2022 by [Chris Nakamura](people/chris-nakamura-40), a former infrastructure engineer who spent years frustrated by the fragmented state of debugging workflows. The company focuses on building unified observability tooling that sits between traditional logging platforms and APM solutions—a niche that's proven surprisingly sticky with mid-sized engineering teams.\n\nThe founding thesis came from Nakamura's experience at larger tech companies where he watched teams cobble together five or six different tools just to trace a single production incident. Meridian's core product aggregates logs, traces, and metrics into what they call a \"narrative view\"—essentially reconstructing the story of what happened in your system without requiring engineers to context-switch between dashboards. Its a deceptively simple idea that turns out to be technically complex to execute well.\n\nFunding came together relatively quickly. [Priya Taylor](people/priya-taylor-85) led the seed round after seeing an early demo, and she brought in [Chris Jackson](people/chris-jackson-91) who had been looking for developer tools plays. [Vera Gonzalez](people/vera-gonzalez-103) joined as a smaller check but has been actively involved in go-to-market strategy. The total seed was $3.2M, closed in late 2022.\n\nOn the advisory side, [Zoe Jackson](people/zoe-jackson-199) has been instrumental in helping Meridian think through enterprise sales motions. Her background in scaling developer-focused products gave the team a playbook they've been iterating on throughout 2023 and into 2024.\n\nMeridian currently has about 14 employees, mostly engineers, operating out of a small office in San Francisco's Dogpatch neighborhood. They've been deliberatley slow on hiring, preferring to keep the team tight while they nail down product-market fit. Revenue numbers aren't public but word is they crossed $500K ARR sometime in early 2024, with a handful of paying customers in the fintech and healthtech spaces.\n\nThe company's biggest challenge right now is differentiation. The observability market is crowded, and larger players like Datadog keep expanding their feature sets. Nakamura has been vocal about staying focused on the \"debugging narrative\" angle rather than trying to become a full platform. Whether that strategy holds as they scale remains to be seen.",
-  "timeline": "- **2022-03-14** | Chris Nakamura incorporates Meridian, begins building initial prototype\n- **2022-08-22** | First demo shown to [Priya Taylor](people/priya-taylor-85), receives positive feedback and term sheet discussions begin\n- **2022-11-03** | Seed round closes at $3.2M with [Chris Jackson](people/chris-jackson-91) and [Vera Gonzalez](people/vera-gonzalez-103) participating\n- **2023-02-17** | Meridian launches private beta, onboards first 12 design partners\n- **2023-06-09** | [Zoe Jackson](people/zoe-jackson-199) joins as formal advisor, begins weekly office hours with team\n- **2023-09-28** | Public launch at a small developer conference in SF, picks up first paying customers\n- **2024-01-15** | Crosses $500K ARR milestone, team celebrates with low-key dinner\n- **2024-05-20** | Hires first dedicated sales rep, begins outbound motion targeting Series B+ startups\n- **2024-11-08** | Ships major \"Narrative 2.0\" update with improved trace visualization\n- **2025-02-14** | Begins early conversations about Series A, [Priya Taylor](people/priya-taylor-85) making introductions to growth-stage funds",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/meridian-40",
-    "name": "Meridian",
-    "category": "startup",
-    "industry": "developer tools",
-    "founded_year": 2022,
-    "founders": [
-      "people/chris-nakamura-40"
-    ],
-    "investors": [
-      "people/priya-taylor-85",
-      "people/chris-jackson-91",
-      "people/vera-gonzalez-103"
-    ],
-    "employees": [
-      "people/kate-kapoor-150"
-    ],
-    "advisors": [
-      "people/zoe-jackson-199"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__meta-2.json b/eval/data/world-v1/companies__meta-2.json
deleted file mode 100644
index 79c7e3c0a..000000000
--- a/eval/data/world-v1/companies__meta-2.json
+++ /dev/null
@@ -1,15 +0,0 @@
-{
-  "slug": "companies/meta-2",
-  "type": "company",
-  "title": "Meta (Cybersecurity)",
-  "compiled_truth": "Meta is a cybersecurity firm founded in 1997, not to be confused with the social media giant of the same name. Operating in the enterprise security space for over two decades, the company has built a reputation as a quiet but effective acquirer of smaller security startups and niche technology providers.\n\nThe company specializes in network security infrastructure and threat detection systems, serving primarily Fortune 500 clients and government contractors. Their flagship product line focuses on perimeter defense and intrusion detection, though they've expanded considerably through strategic acquisitions over the years. Meta's approach has always been to identify promising early-stage cybersecurity companies and integrate their technology into the broader Meta ecosystem.\n\nIn recent years, Meta has been particularly active in the acqusition market, snapping up several AI-driven security startups looking to modernize their offerings. The company completed at least three acquisitions in 2024 alone, focusing on machine learning-based threat analysis and zero-trust architecture providers. Their M&A strategy tends to favor companies with strong technical teams rather than those with large customer bases—they're buying talent and IP, not revenue.\n\nLeadership at Meta Cybersecurity has remained relatively stable, with most of the executive team having been with the company for over a decade. This continuity has allowed them to maintain consistent strategic direction even as the cybersecurity landscape shifts dramatically. They've been rumored to be in discussions with [Anduril Industries](companies/anduril-industries) regarding potential partnership opportunities in the defense sector, though neither party has confirmed these reports.\n\nThe firm maintains a low public profile compared to flashier competitors, preferring to let their client relationships speak for themselves. Their government contracting work, in particular, requires discretion. Meta has also been mentioned in connection with [Palantir Technologies](companies/palantir-technologies) as a potential acquisition target, though industry analysts consider this unlikely given Meta's own acquisition-focused strategy and the cultural differences between the two organizations.\n\nHeadquartered in the Washington D.C. metro area, Meta employs approximately 800 people across their main office and satellite locations in Austin and Tel Aviv.",
-  "timeline": "- **2021-03-15** | Meta acquires small endpoint security startup based in Boston for undisclosed sum\n- **2021-09-22** | Company celebrates 24 years in operation with internal summit featuring keynote on future of zero-trust\n- **2022-04-08** | Meta Cybersecurity signs major contract with Department of Defense for network monitoring services\n- **2022-11-30** | Opens new R&D facility in Tel Aviv focused on threat intelligence\n- **2023-06-14** | Partnership discussions reportedly begin with [Anduril Industries](companies/anduril-industries) around defense applications\n- **2024-02-19** | Completes acquisition of AI security startup, third deal in eight months\n- **2024-08-05** | Meta leadership meets with [Palantir Technologies](companies/palantir-technologies) executives at RSA Conference, sparking merger speculation\n- **2025-01-12** | Launches next-generation threat detection platform incorporating acquired ML technology\n- **2025-07-28** | Announces expansion of Austin office, adding 150 new engineering positions\n- **2026-03-03** | Named to Gartner Magic Quadrant for Enterprise Network Security for fifth consecutive year",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/meta-2",
-    "name": "Meta",
-    "category": "acquirer",
-    "industry": "cybersecurity",
-    "founded_year": 1997
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__microsoft-0.json b/eval/data/world-v1/companies__microsoft-0.json
deleted file mode 100644
index 0630135d3..000000000
--- a/eval/data/world-v1/companies__microsoft-0.json
+++ /dev/null
@@ -1,15 +0,0 @@
-{
-  "slug": "companies/microsoft-0",
-  "type": "company",
-  "title": "Microsoft",
-  "compiled_truth": "Microsoft is a dominant force in the cybersecurity landscape, having transformed itself from a traditional software giant into one of the most aggressive acquirers in the security space. Founded in 1995, the company has methodically built out its security portfolio through strategic acquisitions and internal development, positioning itself as a one-stop shop for enterprise security needs.\n\nThe company's cybersecurity division generates over $20 billion in annual revenue, making it one of the largest security vendors globally. Microsoft's approach has been to embed security deeply into its cloud infrastructure, particularly Azure and Microsoft 365, creating an integrated ecosystem thats difficult for competitors to match. Their Defender suite, Sentinel SIEM platform, and Entra identity solutions form the backbone of security for thousands of enterprises worldwide.\n\nMicrosoft's acquisition strategy has been notably aggressive. They've snapped up numerous startups and established players alike, often integrating the technology directly into their existing platforms. This has created tension with pure-play security vendors who find themselves competing against a company that bundles security features into products their customers already use. Some critics argue this bundling approach leads to \"good enough\" security rather than best-in-class protection, but the convenience factor has proven compelling for many IT departments.\n\nThe company has also invested heavily in threat intelligence, operating one of the largest security research teams in the industry. Their visibility into global attack patterns—derived from telemetry across Windows, Azure, and Office 365—gives them unique insights that feed back into their products. Recent moves have focused on AI-powered security tools, with Microsoft positioning Copilot for Security as a force multiplier for understaffed security teams.\n\nLeadership under Satya Nadella has prioritized security as a core pillar, especially following several high-profile breaches affecting Microsoft's own infrastructure. The company has faced scrutiny from government agencies and enterprise customers demanding better baseline security, prompting internal reorganizations and the Secure Future Initiative. Despite these challanges, Microsoft remains a category-defining player that shapes how the industry thinks about integrated security platforms.",
-  "timeline": "- **2021-03-15** | Microsoft announces acquisition of RiskIQ for threat intelligence capabilities, expanding its external attack surface management\n- **2021-07-22** | Completed purchase of CloudKnox Security to bolster identity and access management portfolio\n- **2022-04-18** | Launched Microsoft Entra brand, consolidating identity products under unified naming\n- **2022-11-09** | Security revenue surpasses $20 billion annually, making MSFT one of the largest security vendors globally\n- **2023-03-28** | Unveiled Security Copilot at Ignite, bringing generative AI to security operations workflows\n- **2023-08-14** | Faced congressional scrutiny following Chinese threat actor breach of government email accounts via compromised signing keys\n- **2024-01-22** | Announced Secure Future Initiative following internal security review, pledging fundamental changes to development practices\n- **2024-06-11** | Expanded partnership with major defense contractors for classified cloud security workloads\n- **2025-02-19** | Acquired endpoint detection startup to enhance Defender capabilities in OT/IoT environments\n- **2025-09-03** | Microsoft Security leadership presented at RSA Conference on next-generation SIEM architecture",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/microsoft-0",
-    "name": "Microsoft",
-    "category": "acquirer",
-    "industry": "cybersecurity",
-    "founded_year": 1995
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__mosaic-14.json b/eval/data/world-v1/companies__mosaic-14.json
deleted file mode 100644
index d779ae426..000000000
--- a/eval/data/world-v1/companies__mosaic-14.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "companies/mosaic-14",
-  "type": "company",
-  "title": "Mosaic - Consumer Social Startup",
-  "compiled_truth": "Mosaic is a consumer social startup founded in 2018 by [Vera Chen](people/vera-chen-14), who serves as the company's CEO. The company operates in the consumer social space, building products that aim to reimagine how people connect and share experiences online. Based on the premise that traditional social media has become too performative and shallow, Mosaic set out to create more authentic digital spaces for meaningful interaction.\n\nThe platform's core product allows users to create collaborative visual stories—essentially shared digital scrapbooks that multiple people can contribute to in real-time. Think of it as a blend between Pinterest boards and group chats, but with richer media capabilities. The name \"Mosaic\" reflects this vision: individual pieces coming together to form something beautiful and cohesive.\n\nVera Chen built the initial prototype while working nights and weekends, drawing on her background in interaction design and her frustration with existing social platforms. Early traction came from college students coordinating group trips and long-distance friend groups trying to stay connected. The organic growth caught the attention of several investors in the Bay Area.\n\n[Helen Martinez](people/helen-martinez-87) led an early investment round, providing crucial capital that allowed Mosaic to expand its engineering team and improve infastructure. Martinez saw potential in Chen's vision and the company's strong retention metrics among its early user base. The investment also brought valuable mentorship to the young founder.\n\nThe company has faced significant competition from established players who've tried to replicate similar features. Instagram's \"Collabs\" and Snapchat's shared stories both emerged after Mosaic gained traction. However, the startup has maintained its niche by focusing on depth over breadth—their users create fewer posts but spend more time on each one.\n\nMosiac currently employs around 35 people, mostly engineers and designers. The team operates with a hybrid work model, with offices in San Francisco. Revenue comes primarily from a freemium subscription model, though the company has experimented with brand partnerships for special templates and features.",
-  "timeline": "- **2018-03-15** | Vera Chen incorporates Mosaic and begins building the first prototype\n- **2018-11-02** | Beta launch to 500 users, mostly from Chen's network and local universities\n- **2019-06-20** | [Helen Martinez](people/helen-martinez-87) leads seed round of $2.1M\n- **2020-01-08** | Mosaic hits 100,000 registered users during pandemic surge in social app usage\n- **2021-04-12** | Series A closes at $12M, company expands engineering team to 20\n- **2022-09-30** | Launch of Mosaic Pro subscription tier with premium collaborative features\n- **2023-03-18** | [Vera Chen](people/vera-chen-14) speaks at SXSW on \"Building for Authentic Connection\"\n- **2024-07-22** | Partnership announced with major photo printing service for physical mosaic books\n- **2025-02-14** | Company reaches 2 million monthly active users milestone\n- **2025-11-03** | Mosaic acquires small AR startup to integrate spatial features into platform",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/mosaic-14",
-    "name": "Mosaic",
-    "category": "startup",
-    "industry": "consumer social",
-    "founded_year": 2018,
-    "founders": [
-      "people/vera-chen-14"
-    ],
-    "investors": [
-      "people/helen-martinez-87"
-    ],
-    "employees": [
-      "people/chris-rodriguez-124"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__nea-13.json b/eval/data/world-v1/companies__nea-13.json
deleted file mode 100644
index 5d7542e78..000000000
--- a/eval/data/world-v1/companies__nea-13.json
+++ /dev/null
@@ -1,14 +0,0 @@
-{
-  "slug": "companies/nea-13",
-  "type": "company",
-  "title": "NEA (New Enterprise Associates)",
-  "compiled_truth": "New Enterprise Associates, commonly known as NEA, stands as one of the largest and most established venture capital firms in the world. Founded in 1977, the firm has grown from its roots in early-stage technology investing to become a multi-stage powerhouse with assets under management exceeding $25 billion. NEA operates across the full spectrum of venture investing, from seed rounds to growth equity, with a particular focus on technology and healthcare sectors.\n\nThe firm maintains offices in Menlo Park, San Francisco, New York, Boston, and internationally, giving it substantial reach across major startup ecosystems. NEA's investment philosophy emphasizes long-term partnerships with founders, and they've backed some of the most consequential companies of the past several decades including Salesforce, Workday, and Uber. Their healthcare practice is particularly notable, having invested in numerous successful biotech and medical device companies.\n\nIn recent years NEA has continued to raise substantial funds, with their latest flagship fund exceeding $3.6 billion. The firm operates with a relatively large partnership compared to some peers, allowing them to cover more ground but sometimes leading to questions about decision-making speed. Partners like Scott Sandell and Peter Barris have shaped the firms direction over multiple decades, though newer partners are increasingly taking lead roles on deals.\n\nNEA has shown interest in emerging areas like AI infrastructure and climate tech, competing with firms like [Andreessen Horowitz](companies/a16z-9) for the hottest deals. Their approach tends to be more traditional than some newer entrants to venture — they're known for thorough due dilligence and sometimes slower processes, which can be both a feature and a bug depending on founder preferences. The firm frequently co-invests alongside other major players including [Sequoia Capital](companies/sequoia-capital-6), particularly on larger growth rounds where syndicate diversity matters to founders.\n\nNEA's brand carries significant weight in boardrooms and with LPs, though they face ongoing pressure to demonstrate continued relevance as the venture landscape evolves rapidly around them.",
-  "timeline": "- **2021-03-15** | NEA closes Fund XIV at $3.6 billion, one of the largest funds in firm history\n- **2021-09-22** | Lead investment in Series B for AI-native cybersecurity startup alongside [Sequoia Capital](companies/sequoia-capital-6)\n- **2022-04-08** | Partner Hannah Kreiswirth promoted to lead healthcare investing practice\n- **2022-11-30** | NEA portfolio company exits via SPAC merger, generating 8x return\n- **2023-06-14** | Announced strategic focus on climate tech, committing $500M to sector\n- **2023-10-02** | Co-led $180M growth round in enterprise AI company with [Andreessen Horowitz](companies/a16z-9)\n- **2024-02-19** | Opened new office in London to expand European presence\n- **2024-08-07** | Scott Sandell announces transition to Chairman role, new managing partners named\n- **2025-01-23** | Led seed round for stealth quantum computing startup at $40M valuation\n- **2025-05-11** | NEA portfolio company IPO on NYSE, largest venture-backed healthcare listing of the year",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/nea-13",
-    "name": "NEA",
-    "category": "vc",
-    "industry": "venture capital"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__nexus-41.json b/eval/data/world-v1/companies__nexus-41.json
deleted file mode 100644
index 51c75f3b5..000000000
--- a/eval/data/world-v1/companies__nexus-41.json
+++ /dev/null
@@ -1,21 +0,0 @@
-{
-  "slug": "companies/nexus-41",
-  "type": "company",
-  "title": "Nexus",
-  "compiled_truth": "Nexus is a biotech startup founded in 2023 by [Alice Kim](people/alice-kim-41), a computational biologist who previously spent nearly a decade at Genentech before striking out on her own. The company operates in the synthetic biology space, specifically focused on developing novel protein engineering platforms that leverage machine learning to accelerate drug discovery timelines.\n\nThe founding thesis behind Nexus centers on a simple but powerful idea: traditional protein design is too slow and too expensive. [Alice Kim](people/alice-kim-41) built the initial prototype while still moonlighting at her previous role, using transformer-based models to predict protein folding outcomes with what she claims is 40% better accuracy than existing tools. Bold claim. The early data seems to back it up, though peer review is still pending on their foundational paper.\n\nNexus raised a $4.2M seed round in late 2023, led by a syndicate of biotech-focused angels and one undisclosed strategic investor rumored to be connected to a major pharma company. The funds went primarily toward buildling out their wet lab capabilities in South San Francisco and hiring a small but senior team of six full-time employees. Alice has been deliberate about keeping the team lean—she's said publicly that she'd rather have five exceptional people than fifteen mediocre ones.\n\nThe company's go-to-market strategy involves partnering with mid-size pharmaceutical companies who lack the in-house ML expertise to build these platforms themselves. Nexus positions itself as a \"co-pilot\" rather than a replacement, which has helped ease concerns about IP ownership and control. Two pilot partnerships were announced in early 2024, though neither partner has been named publicly.\n\nCulturally, Nexus operates with an almost academic intensity. Weekly journal clubs, mandatory documentation of experiments, open internal debates about methodology. Alice brought this ethos from her research days and has made it core to how the company functions. Some employees thrive in this environment; others have found it exhausting. Turnover has been minimal so far, but the company is still young.",
-  "timeline": "- **2023-03-15** | [Alice Kim](people/alice-kim-41) incorporates Nexus as a Delaware C-corp while still employed at Genentech\n- **2023-06-22** | Alice leaves Genentech to work on Nexus full-time; secures initial $500K pre-seed from angel investors\n- **2023-09-08** | Nexus closes $4.2M seed round; announces plans to open South San Francisco wet lab\n- **2023-11-30** | First full-time hire: Dr. Marcus Chen joins as Head of Protein Engineering\n- **2024-01-17** | Wet lab facility becomes operational; first internal experiments begin\n- **2024-04-03** | Nexus announces two unnamed pharmaceutical partnership pilots\n- **2024-07-12** | [Alice Kim](people/alice-kim-41) presents preliminary platform results at SynBioBeta conference\n- **2024-10-25** | Team expands to six FTEs; company moves to larger office space\n- **2025-02-14** | Submits foundational paper on ML-driven protein folding to Nature Methods\n- **2025-06-01** | Series A discussions reportedly underway with multiple tier-1 biotech VCs",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/nexus-41",
-    "name": "Nexus",
-    "category": "startup",
-    "industry": "biotech",
-    "founded_year": 2023,
-    "founders": [
-      "people/alice-kim-41"
-    ],
-    "employees": [
-      "people/eric-park-151"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__nimbus-5.json b/eval/data/world-v1/companies__nimbus-5.json
deleted file mode 100644
index 5b3a98965..000000000
--- a/eval/data/world-v1/companies__nimbus-5.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "companies/nimbus-5",
-  "type": "company",
-  "title": "Nimbus",
-  "compiled_truth": "Nimbus is a climate tech startup founded in early 2025 by [Mia Anderson](people/mia-anderson-5), a serial entrepreneur with a background in atmospheric science and distributed systems. The company is building what they describe as a \"climate intelligence layer\" — essentially a real-time data platform that aggregates satellite imagery, sensor networks, and predictive models to help enterprises and governments make better decisions around carbon accounting, extreme weather preparedness, and supply chain resiliance.\n\nThe founding story is pretty straightforward. Mia had been working on climate modeling tools at a larger company, got frustrated with how slow things moved, and decided to spin out her own thing. She bootstrapped for about three months before bringing on [Noah Nakamura](people/noah-nakamura-182) as an advisor. Noah's been instrumental in shaping their go-to-market strategy, particularly around enterprise sales cycles and pricing architecture.\n\nNimbus operates with a small but focused team — currently around 8 people, mostly engineers with a couple of climate scientists. They've been pretty heads-down on product development, though theyve started doing some early pilots with logistics companies in the Pacific Northwest. The initial use case seems to be helping shipping and freight operations anticipate weather disruptions and reroute proactively.\n\nWhat makes Nimbus interesting is their approach to data fusion. Rather than building their own sensor network from scratch, they're aggregating existing data sources — NOAA feeds, commercial satellite providers, IoT sensors already deployed by clients — and layering their own ML models on top. This keeps their infrastructure costs relatively low while still delivering actionable insights.\n\nThe company hasn't announced any formal funding rounds yet, though rumors suggest they're in conversations with a few climate-focused VCs. Mia Anderson has been intentionally keeping things quiet, preferring to let the product speak for itself before raising. Their advisory relationship with Noah Nakamura gives them some credibility in enterprise circles, which should help when they do decide to go out for capital.",
-  "timeline": "- **2024-09-15** | [Mia Anderson](people/mia-anderson-5) leaves previous role to begin exploring climate intelligence concepts\n- **2025-01-08** | Nimbus officially incorporated in Delaware\n- **2025-02-14** | [Noah Nakamura](people/noah-nakamura-182) joins as advisor, begins weekly strategy sessions\n- **2025-03-22** | First engineering hire made — backend systems specialist from Google\n- **2025-04-10** | Internal alpha of climate data platform completed\n- **2025-05-18** | Pilot program launched with two Pacific Northwest logistics companies\n- **2025-07-02** | Team expands to 8 full-time employees\n- **2025-08-29** | Nimbus presents at Climate Tech Connect conference in Portland\n- **2025-10-15** | Early discussions begin with climate-focused VC firms",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/nimbus-5",
-    "name": "Nimbus",
-    "category": "startup",
-    "industry": "climate tech",
-    "founded_year": 2025,
-    "founders": [
-      "people/mia-anderson-5"
-    ],
-    "employees": [
-      "people/quinten-nakamura-115"
-    ],
-    "advisors": [
-      "people/noah-nakamura-182"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__nimbus-labs-55.json b/eval/data/world-v1/companies__nimbus-labs-55.json
deleted file mode 100644
index 5259da942..000000000
--- a/eval/data/world-v1/companies__nimbus-labs-55.json
+++ /dev/null
@@ -1,21 +0,0 @@
-{
-  "slug": "companies/nimbus-labs-55",
-  "type": "company",
-  "title": "Nimbus Labs",
-  "compiled_truth": "Nimbus Labs is a developer tools startup founded in 2019 by [Vera Kapoor](people/vera-kapoor-55), who previously spent nearly a decade building infrastructure at larger tech companies before striking out on her own. The company focuses on cloud-native debugging and observability tooling, with their flagship product being a distributed tracing platform that's gained significant traction among mid-sized engineering teams.\n\nThe core thesis behind Nimbus is that debugging microservices shouldn't require a PhD in distributed systems. Their approach combines automatic instrumentation with AI-assisted root cause analysis, letting developers pinpoint issues across complex service meshes without manually correlating logs across dozens of services. It's an opinionated take on observability that's rubbed some infrastructure purists the wrong way, but the product's ease of adoption has won over plenty of converts.\n\nVera has been the public face of the company since day one, frequently speaking at conferences about the future of developer experience. She's known for her direct communication style and has built a small but loyal following on technical blogs. Under her leadership, Nimbus Labs has grown from a three-person team working out of a WeWork to roughly 45 employees spread across San Francisco and a small office in Bangalore.\n\nThe company raised a Series A in late 2021 and has been relatively quiet about fundraising since, though rumors of a Series B have circulated. Nimbus competes in a crowded space against established players like Datadog and newer entrants, but they've carved out a niche by focusing specifically on the debugging workflow rather than trying to be an all-in-one platform. Recent product updates have emphasized integration with popular CI/CD pipelines and expanded support for serverless architectures.\n\n[Vera Kapoor](people/vera-kapoor-55) remains CEO and maintains a hands-on role in product decisions, which some investors see as both a strength and potential bottleneck as the company scales. The next year will likely determine whether Nimbus can break out of its current niche or gets aquired by a larger platform player.",
-  "timeline": "- **2019-03-14** | Nimbus Labs incorporated in Delaware; [Vera Kapoor](people/vera-kapoor-55) listed as sole founder and CEO\n- **2019-11-02** | First public beta launched at a small developer meetup in SF; initial feedback was mixed but enthusiastic from early adopters\n- **2021-06-18** | Closed $8.5M Series A led by Baseline Ventures; announced plans to triple engineering headcount\n- **2022-02-10** | Shipped v2.0 of core tracing platform with AI-assisted analysis features\n- **2022-09-23** | Vera Kapoor delivered keynote at DevOpsCon on \"The Death of Manual Debugging\"\n- **2023-04-05** | Opened Bangalore engineering office; hired first international team members\n- **2023-11-30** | Reached 1,000 paying customers milestone; mostly SMB and mid-market\n- **2024-07-12** | Launched serverless support after months of customer requests\n- **2025-01-20** | Rumored acquisition talks with larger observability vendor fell through\n- **2025-08-03** | Announced partnership with major cloud provider for native integration",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/nimbus-labs-55",
-    "name": "Nimbus Labs",
-    "category": "startup",
-    "industry": "developer tools",
-    "founded_year": 2019,
-    "founders": [
-      "people/vera-kapoor-55"
-    ],
-    "employees": [
-      "people/iris-jones-165"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__orbit-42.json b/eval/data/world-v1/companies__orbit-42.json
deleted file mode 100644
index 569076b06..000000000
--- a/eval/data/world-v1/companies__orbit-42.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "companies/orbit-42",
-  "type": "company",
-  "title": "Orbit - Biotech Startup",
-  "compiled_truth": "Orbit is a biotech startup founded in 2021 by [Jack Patel](people/jack-patel-42), focused on developing novel protein engineering platforms for therapeutic applications. The company emerged from Patel's earlier research work and has positioned itself at the intersection of computational biology and wet lab innovation. Based out of the Boston-Cambridge biotech corridor, Orbit has built a lean but ambitious team.\n\nThe company's core technology revolves around machine learning-driven protein design, enabling faster iteration cycles for drug candidates targeting rare genetic disorders. Their proprietary platform, internally called \"Orbital,\" can predict protein folding outcomes with unusual accuracy, cutting development timelines significantly. Early partnerships with academic institutions have validated their approach, though comercial traction remains nascent.\n\nFunding has come from angel investors including [Julia Davis](people/julia-davis-86) and [Zoe Gonzalez](people/zoe-gonzalez-100), both of whom participated in Orbit's seed round. Davis in particular has been an active advisor, leveraging her network in the life sciences space to open doors for the young company. Gonzalez contributed not just capital but also operational guidance, having scaled biotech ventures before.\n\nJack Patel serves as CEO and remains deeply involved in the scientific direction. He's known for being hands-on in the lab despite growing management responsibilites. The team has grown to roughly 15 people as of late 2024, with key hires in protein chemistry and ML engineering.\n\nOrbit has kept a relatively low profile compared to flashier biotech startups, preferring to let results speak. They've published two peer-reviewed papers and presented at major conferences including the Biotech Showcase in San Francisco. The company is currently running preclinical studies for their lead program, OBT-101, targeting a rare metabolic condition. Industry watchers see Orbit as a company to watch—small but technically rigorous, with a founder who understands both the science and the business.",
-  "timeline": "- **2021-03-15** | Orbit incorporated in Delaware by [Jack Patel](people/jack-patel-42)\n- **2021-08-22** | Closed $1.2M seed round led by [Julia Davis](people/julia-davis-86)\n- **2022-02-10** | First version of Orbital platform completed internally\n- **2022-09-18** | Published initial findings in Nature Biotechnology\n- **2023-01-24** | [Zoe Gonzalez](people/zoe-gonzalez-100) joins as advisor and investor\n- **2023-06-30** | Hired Dr. Maria Chen as Head of Protein Chemistry\n- **2024-01-12** | Presented OBT-101 preclinical data at JP Morgan Healthcare Conference\n- **2024-07-08** | Expanded lab space in Cambridge, MA\n- **2025-03-20** | Initiated IND-enabling studies for lead program\n- **2025-11-05** | Announced collaboration with major pharma partner (undisclosed)",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/orbit-42",
-    "name": "Orbit",
-    "category": "startup",
-    "industry": "biotech",
-    "founded_year": 2021,
-    "founders": [
-      "people/jack-patel-42"
-    ],
-    "investors": [
-      "people/julia-davis-86",
-      "people/zoe-gonzalez-100"
-    ],
-    "employees": [
-      "people/rachel-jones-152"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__prism-43.json b/eval/data/world-v1/companies__prism-43.json
deleted file mode 100644
index 2d99ca93b..000000000
--- a/eval/data/world-v1/companies__prism-43.json
+++ /dev/null
@@ -1,31 +0,0 @@
-{
-  "slug": "companies/prism-43",
-  "type": "company",
-  "title": "Prism",
-  "compiled_truth": "Prism is a cybersecurity startup founded in 2023 by [David Patel](people/david-patel-43), who previously spent nearly a decade building threat detection systems at larger security firms. The company focuses on what it calls 'adaptive perimeter defense'—essentially AI-driven intrusion detection that learns an organization's normal traffic patterns and flags anomolies in real-time. Its early traction has been notable, particularly among mid-market financial services companies who find enterprise solutions too expensive but need more than basic firewall protections.\n\nThe founding story is pretty straightforward. David had grown frustrated with the slow pace of innovation at his previous employer and saw an opening in the market for lightweight, intelligent security tooling that didn't require a dedicated SOC team to operate. He bootstrapped the initial prototype over six months before raising a seed round.\n\nPrism's investor syndicate includes [Carol Jackson](people/carol-jackson-81), [Rosa Jackson](people/rosa-jackson-90), and [Tina Hernandez](people/tina-hernandez-97). Carol led the seed round and reportedly pushed hard for the company to focus on the SMB market rather than chasing enterprise deals too early. This strategic direction has shaped much of Prism's go-to-market approach. Rosa came in through an angel allocation and has been relatively hands-off, while Tina joined the cap table in a follow-on extension round in late 2024.\n\nOn the advisory side, [Alice Davis](people/alice-davis-172) provides guidance on product architecture—she's known for her work on distributed systems and has been instrumental in helping Prism scale its detection engine. [Olivia Miller](people/olivia-miller-176) advises on sales strategy and customer success, drawing on her background in enterprise software GTM.\n\nThe team has grown to around 18 people, mostly engineers, with a small but scrappy sales org. Prism operates out of Austin but has several remote employees scattered across the US. The company culture skews technical and moves fast—David himself still reviews most major PRs. Revenue is growing but the company isn't yet profitable, which is typical for this stage. They're expected to raise a Series A sometime in mid-2025.",
-  "timeline": "- **2023-02-14** | David Patel incorporates Prism and begins building initial prototype\n- **2023-07-22** | Seed round closes with [Carol Jackson](people/carol-jackson-81) leading, $2.1M raised\n- **2023-11-03** | First paying customer signs—a regional credit union in Texas\n- **2024-01-18** | [Alice Davis](people/alice-davis-172) joins as technical advisor\n- **2024-04-09** | Prism launches v1.0 of its adaptive perimeter defense platform\n- **2024-08-15** | Team hits 12 employees, opens small Austin office\n- **2024-10-30** | Extension round adds [Tina Hernandez](people/tina-hernandez-97) to investor group\n- **2025-01-22** | [Olivia Miller](people/olivia-miller-176) begins advising on GTM strategy\n- **2025-03-11** | ARR crosses $800K, Series A conversations begin",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/prism-43",
-    "name": "Prism",
-    "category": "startup",
-    "industry": "cybersecurity",
-    "founded_year": 2023,
-    "founders": [
-      "people/david-patel-43"
-    ],
-    "investors": [
-      "people/carol-jackson-81",
-      "people/rosa-jackson-90",
-      "people/tina-hernandez-97"
-    ],
-    "employees": [
-      "people/mia-singh-153"
-    ],
-    "advisors": [
-      "people/alice-davis-172",
-      "people/olivia-miller-176",
-      "people/zoe-jackson-199"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__pulse-8.json b/eval/data/world-v1/companies__pulse-8.json
deleted file mode 100644
index de577071c..000000000
--- a/eval/data/world-v1/companies__pulse-8.json
+++ /dev/null
@@ -1,27 +0,0 @@
-{
-  "slug": "companies/pulse-8",
-  "type": "company",
-  "title": "Pulse - EdTech Startup",
-  "compiled_truth": "Pulse is an edtech startup founded in 2022 by [Yara Johnson](people/yara-johnson-8), a former learning experience designer who spent nearly a decade observing how students actually engage with digital content. The company's core product is a real-time engagement analytics platform designed for K-12 classrooms and higher education institutions. Unlike traditional LMS analytics that track completion rates and grades, Pulse monitors micro-behaviors—pause patterns, scroll velocity, re-reads—to give educators a genuine sense of whether students are struggling before they fail a test.\n\nThe founding thesis came from Johnson's frustration with existing tools that treated engagement as a binary: either a student watched the video or they didn't. Pulse argues that the *how* matters more than the whether. Their proprietary algorithm flags what they call \"confusion signals\" and surfaces them to teachers in a simple dashboard. Early pilots in three school districts showed a 23% reduction in students falling behind, though critics have raised privacy concerns about the level of behavioral tracking involved.\n\nFunding has been modest but strategic. [Eric Martinez](people/eric-martinez-93) led the seed round in late 2022, bringing not just capital but connections to several charter school networks in Texas and California. Martinez has been vocal about his belief that edtech needs more \"unsexy infrastructure\" plays rather than consumer apps, and Pulse fits that thesis perfectly. The company currently employs around 15 people, mostly engineers and former educators.\n\n[David Kim](people/david-kim-186) serves as an advisor, helping Pulse navigate enterprise sales cycles and district procurement processes—notoriously slow and bureacratic. Kim's background in B2B SaaS has been instrumental in shaping Pulse's go-to-market strategy, which prioritizes landing a few large district contracts over chasing individual schools. As of early 2024, Pulse has contracts with 12 districts serving roughly 40,000 students combined. Revenue isn't disclosed but is rumored to be in the low seven figures. Yara Johnson remains CEO and has been clear she's building for the long haul, not a quick exit.",
-  "timeline": "- **2022-03-14** | Yara Johnson incorporates Pulse after leaving her role at a major textbook publisher\n- **2022-09-08** | Closes seed round led by [Eric Martinez](people/eric-martinez-93), raising $1.8M\n- **2022-11-20** | First pilot launches in Austin ISD with 3 middle schools\n- **2023-02-15** | [David Kim](people/david-kim-186) joins as official advisor\n- **2023-06-01** | Pulse ships v2.0 with redesigned teacher dashboard based on pilot feedback\n- **2023-10-12** | Signs first major district contract with Fresno Unified (18,000 students)\n- **2024-01-29** | Presents at SXSWedu panel on ethical student analytics\n- **2024-05-17** | Expands engineering team to 9 people, opens small Denver office\n- **2024-11-03** | Reaches 40,000 students across 12 districts\n- **2025-02-22** | Begins early conversations about Series A with several edtech-focused VCs",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/pulse-8",
-    "name": "Pulse",
-    "category": "startup",
-    "industry": "edtech",
-    "founded_year": 2022,
-    "founders": [
-      "people/yara-johnson-8"
-    ],
-    "investors": [
-      "people/eric-martinez-93"
-    ],
-    "employees": [
-      "people/xavier-nakamura-118"
-    ],
-    "advisors": [
-      "people/david-kim-186"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__pulse-labs-58.json b/eval/data/world-v1/companies__pulse-labs-58.json
deleted file mode 100644
index f8271f22f..000000000
--- a/eval/data/world-v1/companies__pulse-labs-58.json
+++ /dev/null
@@ -1,26 +0,0 @@
-{
-  "slug": "companies/pulse-labs-58",
-  "type": "company",
-  "title": "Pulse Labs",
-  "compiled_truth": "Pulse Labs is a developer tools startup founded in 2019 by [Rachel Lopez](people/rachel-lopez-58), who previously spent nearly a decade building internal tooling at larger tech companies before striking out on her own. The company focuses on API observability and debugging tools, helping engineering teams identify performance bottlenecks and trace issues across distributed systems. Their flagship product, Pulse Trace, has gained traction among mid-sized SaaS companies looking for alternatives to more expensive enterprise solutions.\n\nThe company operates with a relatively lean team of around 35 employees, mostly engineers, spread across San Francisco and a satellite office in Austin. Rachel has been vocal about maintaining a sustainable growth trajectory rather than chasing hypergrowth, which has shaped the company's culture and hiring practices. This philosophy resonated with their investor group, which includes [Carol Jackson](people/carol-jackson-81), [Priya Taylor](people/priya-taylor-85), and [Rosa Jackson](people/rosa-jackson-90).\n\nPulse Labs raised a $4.2M seed round in early 2020, followed by a Series A of $18M in 2022 led by Priya Taylor's fund. The Series A came at a time when developer tooling was seeing significant investor intrest, and Pulse was well-positioned with strong retention metrics among its early customers. Rosa Jackson joined as an angel investor during the seed round and has remained an active advisor, particularly on go-to-market strategy.\n\nRecent moves include expanding their platform to support OpenTelemetry natively, a decision that required significant engineering investment but opened up compatability with a broader ecosystem. The company also launched a free tier in late 2024 aimed at individual developers and small teams, a strategic bet on bottom-up adoption. Rachel Lopez has mentioned in interviews that they're exploring AI-assisted debugging features, though nothing concrete has been announced yet.\n\nPulse Labs competes with established players like Datadog and newer entrants in the observability space, but differentiates through pricing transparency and a focus on developer experience over enterprise feature bloat.",
-  "timeline": "- **2019-03-15** | Pulse Labs incorporated by [Rachel Lopez](people/rachel-lopez-58) in Delaware\n- **2020-01-22** | Closed $4.2M seed round with participation from [Rosa Jackson](people/rosa-jackson-90)\n- **2020-09-08** | Launched Pulse Trace beta to first 50 customers\n- **2021-06-14** | Reached 200 paying customers milestone\n- **2022-04-03** | Announced $18M Series A led by [Priya Taylor](people/priya-taylor-85)\n- **2022-11-17** | Opened Austin office, hired VP of Engineering\n- **2023-05-22** | Rachel Lopez spoke at DevToolsCon on sustainable startup growth\n- **2024-02-09** | Shipped native OpenTelemetry support in Pulse Trace 3.0\n- **2024-10-30** | Launched free tier for individual developers\n- **2025-03-12** | [Carol Jackson](people/carol-jackson-81) joined board as observer seat",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/pulse-labs-58",
-    "name": "Pulse Labs",
-    "category": "startup",
-    "industry": "developer tools",
-    "founded_year": 2019,
-    "founders": [
-      "people/rachel-lopez-58"
-    ],
-    "investors": [
-      "people/carol-jackson-81",
-      "people/priya-taylor-85",
-      "people/rosa-jackson-90"
-    ],
-    "employees": [
-      "people/alice-jones-168"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__quantum-7.json b/eval/data/world-v1/companies__quantum-7.json
deleted file mode 100644
index 3ce99532f..000000000
--- a/eval/data/world-v1/companies__quantum-7.json
+++ /dev/null
@@ -1,27 +0,0 @@
-{
-  "slug": "companies/quantum-7",
-  "type": "company",
-  "title": "Quantum",
-  "compiled_truth": "Quantum is a fintech startup founded in 2022 by [Ulrich Johnson](people/ulrich-johnson-7), a serial entrepreneur with a background in quantitative finance and distributed systems. The company emerged from Johnson's frustration with the sluggish settlement times and opaque fee structures that plague traditional payment rails. Based out of Austin, Texas, Quantum has built a real-time payment reconciliation platform targeting mid-market e-commerce businesses and SaaS companies.\n\nThe core product offers instant transaction matching, automated dispute resolution, and predictive cash flow analytics. What sets Quantum apart from competitors is their proprietary matching algorithm, which reportedly achieves 99.7% accuracy on first-pass reconciliation—a significant improvement over industry standards. The platform integrates with major payment processors, banking APIs, and accounting software, positioning itself as the connective tissue in a fragmented fintech ecosystem.\n\nEarly backing came from [Kate Anderson](people/kate-anderson-107), who led a $2.1M seed round in late 2022. Anderson's involvement brought not just capital but credibility, given her track record of identifying breakout fintech plays. The company has since grown to around 25 employees, with plans to double headcount by end of 2025.\n\nOn the advisory side, [Noah Williams](people/noah-williams-198) has been instrumental in shaping Quantum's go-to-market strategy. Williams' connections in the enterprise software space have opened doors to several pilot programs with Fortune 500 companies—a surprising feat for such a young startup. His guidance on pricing and packaging helped the team move away from a pure usage-based model toward a hybrid subscription approach that's proven more predictable for customers and investors alike.\n\nQuantum's roadmap includes international expansion, starting with the UK and EU markets where PSD2 regulations have created fertile ground for innovative payment solutions. There's also talk of an AI-powered fraud detection layer, though details remain sparse. The company operates somewhat stealthily, preferring to let product traction speak rather than chasing press coverage.",
-  "timeline": "- **2022-03-14** | Ulrich Johnson incorporates Quantum in Delaware, begins recruiting founding engineering team\n- **2022-09-22** | Closes $2.1M seed round led by [Kate Anderson](people/kate-anderson-107)\n- **2022-11-08** | Launches private beta with 12 e-commerce customers\n- **2023-02-15** | [Noah Williams](people/noah-williams-198) joins as lead advisor, focuses on GTM stratgy\n- **2023-06-30** | Exits beta, announces general availability of reconciliation platform\n- **2023-10-12** | Surpasses 200 paying customers, hits $1M ARR milestone\n- **2024-04-18** | Opens Austin headquarters, team grows to 25 employees\n- **2024-08-07** | Begins enterprise pilot program with two Fortune 500 retailers\n- **2025-01-20** | Announces plans for UK expansion, begins regulatory groundwork\n- **2025-05-11** | [Ulrich Johnson](people/ulrich-johnson-7) speaks at FinTech Connect conference on real-time reconciliation",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/quantum-7",
-    "name": "Quantum",
-    "category": "startup",
-    "industry": "fintech",
-    "founded_year": 2022,
-    "founders": [
-      "people/ulrich-johnson-7"
-    ],
-    "investors": [
-      "people/kate-anderson-107"
-    ],
-    "employees": [
-      "people/tina-lopez-117"
-    ],
-    "advisors": [
-      "people/noah-williams-198"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__quantum-labs-57.json b/eval/data/world-v1/companies__quantum-labs-57.json
deleted file mode 100644
index 589236ea5..000000000
--- a/eval/data/world-v1/companies__quantum-labs-57.json
+++ /dev/null
@@ -1,28 +0,0 @@
-{
-  "slug": "companies/quantum-labs-57",
-  "type": "company",
-  "title": "Quantum Labs",
-  "compiled_truth": "Quantum Labs is an early-stage biotech startup founded in 2024 by [Liam Wilson](people/liam-wilson-57), a computational biologist who previously led protein folding research at a major pharma company. The company operates out of a small lab space in Cambridge, MA, though much of the early work has been computational in nature.\n\nThe startup focuses on quantum computing applications for drug discovery, specifically targeting protein-ligand binding simulations that would take classical computers years to process. Their core thesis is that near-term quantum hardware, combined with clever error mitigation techniques, can already provide meaningful speedups for certain molecular dynamics calculations. Its a bold bet, and not everyone in the industry is convinced the hardware is ready.\n\nQuantum Labs raised a pre-seed round in late 2024, with [Rachel Brown](people/rachel-brown-95) leading the investment. Rachel has been particularly bullish on quantum-adjacent biotech plays and saw Liam's background as uniquely suited to bridge the gap between quantum computing hype and actual pharmaceutical applications. [Rosa Miller](people/rosa-miller-98) also participated in the round, bringing her experience scaling deep tech companies.\n\nOn the advisory side, the company brought on [Tara Johnson](people/tara-johnson-189) to help navigate regulatory pathways and partnership discussions with larger pharma players. Tara's connections have already opened doors to several exploratory conversations, though nothing has been announced publically yet.\n\nThe team remains small—just five people including Liam—but they've made progress on their initial benchmarking studies. Early results suggest their hybrid classical-quantum approach can reduce simulation time by roughly 40% for certain small molecule interactions. Whether this translates to real-world drug discovery value remains to be seen. Quantum Labs is currently focused on publishing these findings to establish credibility before pursuing a larger seed round, likely in mid-2025.",
-  "timeline": "- **2024-01-15** | Liam Wilson begins preliminary research and files initial IP for quantum-enhanced molecular simulation methods\n- **2024-03-22** | Quantum Labs officially incorporated in Delaware\n- **2024-05-10** | [Rachel Brown](people/rachel-brown-95) commits to leading pre-seed investment after initial pitch\n- **2024-06-18** | Lab space secured in Cambridge, MA; first equipment purchases made\n- **2024-07-30** | [Rosa Miller](people/rosa-miller-98) joins the round, bringing total pre-seed to $1.8M\n- **2024-09-12** | [Tara Johnson](people/tara-johnson-189) formally joins as advisor\n- **2024-11-05** | First proof-of-concept results show promising speedups on protein-ligand simulations\n- **2025-01-20** | Team expands to five with hire of quantum software engineer from IBM\n- **2025-03-08** | Submits first paper to Nature Computational Science on hybrid simulation methodology",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/quantum-labs-57",
-    "name": "Quantum Labs",
-    "category": "startup",
-    "industry": "biotech",
-    "founded_year": 2024,
-    "founders": [
-      "people/liam-wilson-57"
-    ],
-    "investors": [
-      "people/rachel-brown-95",
-      "people/rosa-miller-98"
-    ],
-    "employees": [
-      "people/frank-moore-167"
-    ],
-    "advisors": [
-      "people/tara-johnson-189"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__quasar-44.json b/eval/data/world-v1/companies__quasar-44.json
deleted file mode 100644
index 4598cedeb..000000000
--- a/eval/data/world-v1/companies__quasar-44.json
+++ /dev/null
@@ -1,27 +0,0 @@
-{
-  "slug": "companies/quasar-44",
-  "type": "company",
-  "title": "Quasar",
-  "compiled_truth": "Quasar is a data infrastructure startup founded in early 2025 by [Mark Wilson](people/mark-wilson-44), a serial entrepreneur with deep roots in distributed systems. The company emerged from Wilson's frustration with existing data pipeline tools, which he found too brittle for modern real-time workloads. Based in San Francisco, Quasar is building what they call a \"unified data fabric\" — essentially a layer that sits between data sources and downstream applications, handling ingestion, transformation, and delivery with minimal configuration.\n\nThe founding team is lean but experienced. Mark Wilson previously led infrastructure at two mid-stage startups, one of wich was aquired by Snowflake in 2022. He's known for strong opinions on developer experience and has been vocal on Twitter about what he sees as the over-complexity of the modern data stack. Early angel investment came from [Jack Davis](people/jack-davis-89), who reportedly wrote a check after a single demo meeting. Davis has been an active advisor beyond just capital, making introductions to potential design partners in the fintech space.\n\nOn the advisory side, Quasar brought on [Grace Singh](people/grace-singh-197) to help shape go-to-market strategy. Singh's background in enterprise sales has already influenced how the company thinks about pricing and packaging. Internal docs suggest they're leaning toward a consumption-based model with a generous free tier to drive adoption among smaller teams.\n\nQuasar is still in stealth mode as of mid-2025, though they've been quietly onboarding design partners. Early feedback has centered on the product's speed — some users report 10x improvements in query latency compared to legacy tools. The tech stack is Rust-heavy, which aligns with Wilson's preference for performance-first engineering. There's some chatter that a seed round is in the works, though nothing confirmed publicly. The company employs around eight people, mostly engineers recruited from Wilson's network.",
-  "timeline": "- **2025-01-14** | Quasar incorporated in Delaware by [Mark Wilson](people/mark-wilson-44)\n- **2025-01-28** | Initial angel check from [Jack Davis](people/jack-davis-89), terms undisclosed\n- **2025-02-10** | First engineering hire joins from Databricks\n- **2025-03-05** | [Grace Singh](people/grace-singh-197) formally joins as advisor\n- **2025-03-22** | Internal alpha of core data fabric released to team\n- **2025-04-18** | First design partner signed — a Series B fintech in NYC\n- **2025-05-09** | Wilson presents at private invite-only infrastructure meetup\n- **2025-06-01** | Team grows to eight full-time employees\n- **2025-06-15** | Second design partner onboarded, early latency benchmarks shared internally",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/quasar-44",
-    "name": "Quasar",
-    "category": "startup",
-    "industry": "data infrastructure",
-    "founded_year": 2025,
-    "founders": [
-      "people/mark-wilson-44"
-    ],
-    "investors": [
-      "people/jack-davis-89"
-    ],
-    "employees": [
-      "people/liam-patel-154"
-    ],
-    "advisors": [
-      "people/grace-singh-197"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__ranger-22.json b/eval/data/world-v1/companies__ranger-22.json
deleted file mode 100644
index 802b4627c..000000000
--- a/eval/data/world-v1/companies__ranger-22.json
+++ /dev/null
@@ -1,28 +0,0 @@
-{
-  "slug": "companies/ranger-22",
-  "type": "company",
-  "title": "Ranger",
-  "compiled_truth": "Ranger is a health tech startup founded in 2024 by [Quinten Rodriguez](people/quinten-rodriguez-22), a first-time founder who previously spent six years in clinical operations at major hospital systems. The company is building what it describes as a \"proactive health monitoring platform\" — essentially a combination of wearable integration, predictive analytics, and care coordination tools aimed at catching health issues before they become emergencies.\n\nThe core product pulls data from consumer wearables and runs it through proprietary algorithms that flag concerning patterns. When something looks off, Ranger connects users directly with healthcare providers through an integrated telehealth layer. It's ambitious, maybe overly so for such an early-stage company, but the team seems to be executing well so far.\n\nRanger raised a pre-seed round in early 2024 with participation from [Helen Martinez](people/helen-martinez-87) and [Sarah Wang](people/sarah-wang-104), both of whom have been active in the health tech space. The round was reportedly around $2.1M, though the company hasn't confirmed exact figures publicly. [Beth Williams](people/beth-williams-177) came on as an advisor shortly after, bringing regulatory expertise that will likely prove critical as Ranger navigates FDA considerations around its predictive features.\n\nThe founding team is still small — just seven people as of late 2024 — but they've been hiring aggresively for ML engineering roles. Quinten has been vocal about wanting to build the technical foundation right before scaling the team further. Smart approach, though it means they're moving slower on go-to-market than some competitors.\n\nRanger's initial focus is on cardiovascular health monitoring for adults over 50, a demographic that's both high-risk and increasingly comfortable with wearable technology. Early pilot programs with two regional health systems have shown promising engagement numbers, though clinical outcomes data is still being collected. The company faces stiff competiton from both established players and well-funded startups, but their emphasis on provider integration rather than direct-to-consumer sales could be a meaningful differentiator.",
-  "timeline": "- **2024-01-15** | Ranger incorporated in Delaware by [Quinten Rodriguez](people/quinten-rodriguez-22)\n- **2024-03-08** | Closed pre-seed round with [Helen Martinez](people/helen-martinez-87) and [Sarah Wang](people/sarah-wang-104) participating\n- **2024-04-22** | [Beth Williams](people/beth-williams-177) joins as regulatory advisor\n- **2024-06-10** | First engineering hire — ML lead recruited from Apple Health team\n- **2024-08-14** | Launched private beta with 200 users in Austin area\n- **2024-10-03** | Announced pilot partnership with Memorial Regional Health System\n- **2024-11-19** | Quinten presented at Digital Health Summit on predictive monitoring\n- **2025-02-01** | Second pilot program launched with Coastal Medical Group\n- **2025-04-28** | Team expanded to 12 people, opened small office in Austin\n- **2025-07-15** | Began conversations with FDA around De Novo pathway for predictive features",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/ranger-22",
-    "name": "Ranger",
-    "category": "startup",
-    "industry": "health tech",
-    "founded_year": 2024,
-    "founders": [
-      "people/quinten-rodriguez-22"
-    ],
-    "investors": [
-      "people/helen-martinez-87",
-      "people/sarah-wang-104"
-    ],
-    "employees": [
-      "people/rachel-miller-132"
-    ],
-    "advisors": [
-      "people/beth-williams-177"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__resonance-45.json b/eval/data/world-v1/companies__resonance-45.json
deleted file mode 100644
index b6c7d7e9f..000000000
--- a/eval/data/world-v1/companies__resonance-45.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "companies/resonance-45",
-  "type": "company",
-  "title": "Resonance",
-  "compiled_truth": "Resonance is an enterprise SaaS startup founded in 2022 by [Grace Thomas](people/grace-thomas-45), a former product lead who spent nearly a decade building internal tools at large tech companies before striking out on her own. The company focuses on helping mid-market enterprises manage and optimize their internal communication workflows—think of it as a layer that sits atop Slack, Teams, and email to surface what actually matters and reduce notification fatigue.\n\nThe core product uses machine learning to prioritize messages, flag action items, and generate daily digests tailored to each employee's role and responsiblities. Early customers have described it as \"finally making enterprise chat usable again.\" Resonance has found particular traction in professional services firms and fast-growing startups where information overload is a constant complaint.\n\nGrace Thomas serves as CEO and has been the public face of the company, frequently speaking at SaaS conferences about the hidden costs of context-switching. She's known for her direct communication style and her insistence on dogfooding—the entire Resonance team uses an internal build of the product daily, often shipping fixes within hours of discovering friction points.\n\nThe company has benefited from the guidance of [Yara Singh](people/yara-singh-195), who joined as an advisor shortly after launch. Yara's experience scaling go-to-market motions has been instrumental in shaping Resonance's sales strategy, particularly around land-and-expand deals with departmental buyers. Under her mentorship, the startup has refined its pricing model and built out a small but effective sales team.\n\nResonance operates with a lean team of about 15 people, mostly engineers and a handful of customer success managers. The company is headquartered in Austin but operates fully remote, drawing talent from across North America. Recent product updates have focused on deeper integrations with project managment tools and improved analytics dashboards for IT admins. The roadmap hints at AI-generated meeting summaries and automated escalation paths, though those features remain in beta.",
-  "timeline": "- **2022-03-14** | Resonance incorporated in Delaware by Grace Thomas\n- **2022-06-01** | Closed a $1.8M pre-seed round led by several angels\n- **2022-09-20** | [Yara Singh](people/yara-singh-195) joins as formal advisor\n- **2023-01-11** | Launched private beta with 12 design partners\n- **2023-05-03** | Public launch of Resonance v1.0 with Slack and Teams integrations\n- **2023-08-15** | Reached $500K ARR milestone\n- **2024-02-22** | [Grace Thomas](people/grace-thomas-45) speaks at SaaStr Annual on reducing enterprise noise\n- **2024-07-09** | Shipped analytics dashboard for IT administrators\n- **2025-01-18** | Announced partnership with a major consulting firm for pilot deployment\n- **2025-04-30** | Beta launch of AI meeting summary feature",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/resonance-45",
-    "name": "Resonance",
-    "category": "startup",
-    "industry": "enterprise SaaS",
-    "founded_year": 2022,
-    "founders": [
-      "people/grace-thomas-45"
-    ],
-    "employees": [
-      "people/eric-singh-155"
-    ],
-    "advisors": [
-      "people/yara-singh-195"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__sentinel-23.json b/eval/data/world-v1/companies__sentinel-23.json
deleted file mode 100644
index 5a7b5aca0..000000000
--- a/eval/data/world-v1/companies__sentinel-23.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "companies/sentinel-23",
-  "type": "company",
-  "title": "Sentinel",
-  "compiled_truth": "Sentinel is a consumer social startup founded in 2019 by [Paul Anderson](people/paul-anderson-23), who previously worked in product roles at several mid-stage companies before striking out on his own. The company operates in the increasingly crowded social space, though it's carved out a niche focused on what Anderson calls \"intentional social\" — essentially tools that help people maintain closer relationships with smaller circles rather than broadcasting to large audiences.\n\nThe core product is a mobile app that combines private group messaging with shared memory features. Users can create small groups (capped at 12 people) and the app automatically surfaces shared photos, past conversations, and anniversary reminders. It's not trying to compete with Instagram or TikTok — more like a utility for your actual close friends. The company has been deliberatly slow in scaling, preferring organic growth over paid acquisition.\n\nSentinel raised a seed round in late 2020, though exact figures haven't been disclosed publicly. The team remains small, hovering around 15 employees as of early 2024. [Julia Chen](people/julia-chen-181) serves as an advisor to the company, bringing her expertise in consumer product development and growth strategy. Her involvement reportedly began through a warm intro from a mutual investor.\n\nRecent moves include a pivot toward integrating AI-powered features — specifically, an assistant that helps users remember important dates and suggests conversation starters based on past interactions. Some users have praised this as genuinely useful; others find it slightly creepy. The company has been testing these features in closed beta since mid-2023.\n\nAnderson has been vocal about building a sustainable business rather than chasing hypergrowth. In interviews, he's mentioned that Sentinel may eventually pursue a subscription model rather than advertising, citing concerns about ad-driven incentives corrupting the product's core mission. Whether this philosophy can survive contact with investor expectations remains to be seen. The startup has mostly stayed under the radar, which seems intentional.",
-  "timeline": "- **2019-03-15** | Sentinel incorporated in Delaware by Paul Anderson\n- **2019-09-02** | First prototype launched to 50 beta users\n- **2020-11-18** | Closed seed funding round, terms undisclosed\n- **2021-06-07** | [Julia Chen](people/julia-chen-181) joined as formal advisor\n- **2022-02-14** | Crossed 100,000 registered users milestone\n- **2022-10-03** | Launched group memory feature called \"Moments\"\n- **2023-05-22** | [Paul Anderson](people/paul-anderson-23) spoke at Consumer Social Summit in SF\n- **2023-08-30** | Began closed beta for AI assistant features\n- **2024-01-12** | Expanded engineering team with three new hires\n- **2025-04-08** | Announced partnership with undisclosed messaging platform",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/sentinel-23",
-    "name": "Sentinel",
-    "category": "startup",
-    "industry": "consumer social",
-    "founded_year": 2019,
-    "founders": [
-      "people/paul-anderson-23"
-    ],
-    "employees": [
-      "people/rosa-wilson-133"
-    ],
-    "advisors": [
-      "people/julia-chen-181"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__sequoia-capital-1.json b/eval/data/world-v1/companies__sequoia-capital-1.json
deleted file mode 100644
index 4e4e8b698..000000000
--- a/eval/data/world-v1/companies__sequoia-capital-1.json
+++ /dev/null
@@ -1,14 +0,0 @@
-{
-  "slug": "companies/sequoia-capital-1",
-  "type": "company",
-  "title": "Sequoia Capital",
-  "compiled_truth": "Sequoia Capital stands as one of the most legendary venture capital firms in Silicon Valley history, having backed companies that collectively represent trillions of dollars in market value. Founded in 1972 by Don Valentine, the firm has maintained its position at the apex of the VC world for over five decades. Their portfolio reads like a who's who of tech giants: Apple, Google, Cisco, Oracle, YouTube, Instagram, WhatsApp, and more recently Stripe and Airbnb.\n\nThe firm operates with a philosophy that emphasizes partnering with \"the crazies\" — founders with audacious visions who refuse to accept conventional wisdom. This approach has served them remarkably well, though it hasn't been without its spectacular failures. The FTX debacle in 2022 forced Sequoia to write down a $150 million investment to zero, a rare and very public miss that prompted some internal reflection on due dilligence processes.\n\nIn recent years, Sequoia has undergone significant structural changes. In 2021, they announced a radical restructuring that would transform the firm into a single registered investment adviser, allowing them to hold public stock positions indefinitely rather than distributing shares to LPs after IPOs. This was later partially reversed in 2023 when they split off their China and India operations into seperate entities — a move driven by geopolitical tensions and LP pressure.\n\nThe firm's current leadership includes Roelof Botha as the global managing partner, having taken over from Doug Leone. Botha, who previously served as CFO of PayPal, has been instrumental in deals involving companies like Unity and MongoDB. Their partnership extends across multiple stages, from their scout program to growth-stage investments.\n\nSequoia's relationship with firms like [Andreessen Horowitz](companies/andreessen-horowitz) has been characterized by both competition and mutual respect — they've co-invested on numerous deals while also fiercely competing for the best founders. The firm continues to be a dominant force in AI investing, having backed companies working with partners at [Y Combinator](companies/y-combinator) and other top accelerators. Their AI fund, launched in 2023, demonstrates their commitment to staying at the frontier of technological change.",
-  "timeline": "- **2021-06-15** | Sequoia announces radical restructuring into single permanent fund structure, shocking the VC industry\n- **2022-01-20** | Led $500M Series C round for AI startup alongside [Andreessen Horowitz](companies/andreessen-horowitz)\n- **2022-11-11** | Published memo to portfolio companies following FTX collapse, writing investment down to zero\n- **2023-03-08** | Roelof Botha promoted to sole global managing partner\n- **2023-06-22** | Announced separation of China and India/SEA operations into independent entities\n- **2024-02-14** | Closed new $2.5B early-stage fund focused on AI and climate tech\n- **2024-09-30** | Participated in seed round for [Y Combinator](companies/y-combinator) batch company building developer tools\n- **2025-01-18** | Hosted annual Base Camp event for seed-stage founders in Woodside\n- **2025-07-22** | Published influential research report on AI agent infrastructure opportunities\n- **2026-03-05** | Led $800M growth round for autonomous systems company at $12B valuation",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/sequoia-capital-1",
-    "name": "Sequoia Capital",
-    "category": "vc",
-    "industry": "venture capital"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__spire-46.json b/eval/data/world-v1/companies__spire-46.json
deleted file mode 100644
index bad6040a7..000000000
--- a/eval/data/world-v1/companies__spire-46.json
+++ /dev/null
@@ -1,30 +0,0 @@
-{
-  "slug": "companies/spire-46",
-  "type": "company",
-  "title": "Spire",
-  "compiled_truth": "Spire is a biotech startup founded in 2018 by [Linda Miller](people/linda-miller-46), a veteran researcher with deep expertise in synthetic biology and metabolic engineering. The company operates out of the Boston-Cambridge biotech corridor, where it has quietly built a reputation for innovative approaches to protein therapeutics. Unlike many flashier competitors, Spire has maintained a relatively low profile, preferring to let its science speak for itself.\n\nThe company's core technology platform focuses on engineered protein scaffolds that can be customized for various therapeutic applications, including oncology and rare genetic disorders. Their lead candidate, SPR-201, is currently in Phase I clinical trials for a rare metabolic condition affecting pediatric patients. Early data has been promising, though the team remains cautious about over-hyping preliminary results.\n\nSpire has attracted a notable group of investors including [Wendy Hernandez](people/wendy-hernandez-80), [Eric Martinez](people/eric-martinez-93), and [Rosa Nakamura](people/rosa-nakamura-94). The Series A round closed in late 2020, with subsequent bridge financing helping extend runway through the expensive clinical development phase. The company has been judicious with capital, maintaining a lean team of around 35 employees while outsourcing certain manufacturing and regulatory functions.\n\nOn the advisory side, Spire benefits from guidance from [David Kim](people/david-kim-186) and [Grace Singh](people/grace-singh-197), both of whom bring significant industry experiance to the table. David in particular has been instrumental in shaping the clinical strategy, drawing on his background in rare disease drug development.\n\nLinda Miller continues to serve as CEO, a somewhat unusual arrangement in biotech where scientific founders often transition to CSO roles as companies mature. However, her combination of scientific credibility and business acumen has made the dual role work. She's known for being intensley focused on execution and has built a culture that prioritizes rigor over hype.\n\nRecent months have seen Spire expanding its pipeline discussions with potential pharma partners, though nothing has been announced publicly. The company is also exploring applications of its platform technology in areas beyond its initial therapeutic focus, potentially setting up multiple shots on goal as it matures.",
-  "timeline": "- **2018-03-15** | Spire incorporated in Delaware by [Linda Miller](people/linda-miller-46), initial seed funding from angel investors\n- **2019-08-22** | Published landmark paper in Nature Biotechnology on novel protein scaffold approach\n- **2020-11-30** | Closed $28M Series A led by [Wendy Hernandez](people/wendy-hernandez-80) and [Eric Martinez](people/eric-martinez-93)\n- **2021-06-14** | [David Kim](people/david-kim-186) joins advisory board to help shape clinical development strategy\n- **2022-01-09** | SPR-201 receives FDA orphan drug designation for rare metabolic disorder\n- **2022-09-03** | Expanded lab facilities in Cambridge, added 12 new research positions\n- **2023-04-18** | IND application submitted for SPR-201, cleared by FDA within 30 days\n- **2024-02-11** | First patient dosed in Phase I trial for SPR-201\n- **2024-10-25** | [Rosa Nakamura](people/rosa-nakamura-94) participates in $15M bridge financing round\n- **2025-03-07** | Presented interim Phase I safety data at rare disease conference, well received by analysts",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/spire-46",
-    "name": "Spire",
-    "category": "startup",
-    "industry": "biotech",
-    "founded_year": 2018,
-    "founders": [
-      "people/linda-miller-46"
-    ],
-    "investors": [
-      "people/wendy-hernandez-80",
-      "people/eric-martinez-93",
-      "people/rosa-nakamura-94"
-    ],
-    "employees": [
-      "people/will-kapoor-156"
-    ],
-    "advisors": [
-      "people/david-kim-186",
-      "people/grace-singh-197"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__talon-47.json b/eval/data/world-v1/companies__talon-47.json
deleted file mode 100644
index 429ab8200..000000000
--- a/eval/data/world-v1/companies__talon-47.json
+++ /dev/null
@@ -1,28 +0,0 @@
-{
-  "slug": "companies/talon-47",
-  "type": "company",
-  "title": "Talon",
-  "compiled_truth": "Talon is an edtech startup founded in 2020 by [Diana Thomas](people/diana-thomas-47), who saw an opportunity to reimagine how students engage with technical curriculum. The company's core product is an adaptive learning platform that uses machine learning to personalize coding education pathways for university students and bootcamp participants. Based in Austin, Texas, Talon has quietly built a reputation for its unusually high completion rates—reportedly 3x the industry average for online technical courses.\n\nThe founding story is straightforward. Diana had spent years frustrated by one-size-fits-all approaches to teaching programming. She bootstrapped the initial prototype while still working her day job, then went full-time in late 2020. Early traction came from partnerships with two regional coding bootcamps who were desperate for better retention tools. Word spread.\n\nInvestment came from [Sarah Williams](people/sarah-williams-92), who led a seed round in early 2022. Sarah's background in workforce development made her a natural fit, and she's remained actively involved in shaping Talon's go-to-market strategy. The company has since expanded to serve over 40 educational institutions, with particular strenght in community colleges looking to modernize their CS programs.\n\nOn the advisory side, [David Kim](people/david-kim-186) provides guidance on enterprise sales cycles, having scaled several B2B edtech companies himself. [Noah Williams](people/noah-williams-198) advises on curriculum design and learning science—his academic background complements Diana's more technical instincts. The advisory board meets quarterly, though informal check-ins happen more frequently.\n\nTalon's recent focus has been on expanding beyond pure coding education into adjacent technical skills: data literacy, basic cloud infrastructure, that sort of thing. There's also been internal discusson about whether to pursue K-12 markets, though Diana has been hesitant to dilute focus. The team remains lean at around 25 employees, mostly engineers and instructional designers. Revenue figures aren't public but insiders suggest ARR crossed $2M sometime in 2024.",
-  "timeline": "- **2020-06-15** | Diana Thomas incorporates Talon and begins building MVP\n- **2020-11-02** | First pilot partnership signed with Austin Coding Academy\n- **2022-02-18** | Seed round closes, led by [Sarah Williams](people/sarah-williams-92)\n- **2022-09-10** | [David Kim](people/david-kim-186) joins as formal advisor\n- **2023-03-22** | Talon platform launches publicly, signs 12 institutions in first quarter\n- **2023-08-14** | [Noah Williams](people/noah-williams-198) brought on to advise on learning science\n- **2024-01-29** | Company hits 40 institutional customers milestone\n- **2024-06-05** | Diana presents at ASU+GSV Summit on adaptive learning\n- **2025-02-11** | Talon announces expansion into data literacy curriculum\n- **2025-09-03** | Strategic partnership discussions begin with major community college system",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/talon-47",
-    "name": "Talon",
-    "category": "startup",
-    "industry": "edtech",
-    "founded_year": 2020,
-    "founders": [
-      "people/diana-thomas-47"
-    ],
-    "investors": [
-      "people/sarah-williams-92"
-    ],
-    "employees": [
-      "people/rachel-thomas-157"
-    ],
-    "advisors": [
-      "people/david-kim-186",
-      "people/noah-williams-198"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__tempo-24.json b/eval/data/world-v1/companies__tempo-24.json
deleted file mode 100644
index 10a22f10f..000000000
--- a/eval/data/world-v1/companies__tempo-24.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "companies/tempo-24",
-  "type": "company",
-  "title": "Tempo",
-  "compiled_truth": "Tempo is a biotech startup founded in 2020 by [Quinten Lee](people/quinten-lee-24), focused on developing novel approaches to metabolic disease therapeutics. The company emerged from Lee's frustration with the slow pace of traditional drug discovery and his belief that computational biology could dramatically accelerate the identification of viable drug candidates.\n\nThe company operates out of a modest lab space in South San Francisco, though they've been reportedly looking at expanding into a larger facility given recent growth. Tempo's core platform combines machine learning with high-throughput screening to identify small molecule compounds that modulate metabolic pathways. Their initial focus has been on type 2 diabetes and obesity, though internal documents suggest they're exploring applications in fatty liver disease as well.\n\n[Yara Moore](people/yara-moore-174) serves as an advisor to the company, bringing her extensive experience in regulatory affairs and clinical development strategy. Her involvement has been particularly valuable as Tempo prepares for eventual IND-enabling studies. Moore's connections within the FDA have reportedly helped the team think more strategically about their development timeline.\n\nThe startup has remained relatively quiet compared to other biotech players in the metabolic space, preferring to let data speak rather than hype. Quinten has been deliberate about this approach, often saying in interviews that \"biotech has too much vaporware.\" This philosophy has attracted a certain type of investor—those who prefer substance over flash.\n\nTempo raised a seed round in early 2021 and has since closed a Series A, though exact figures haven't been publicly disclosed. The team has grown to approximately 25 people, mostly bench scientists and computational biologists. They've published a couple of papers in mid-tier journals, nothing splashy, but the work demonstrates a solid methodological foundation. Recent rumors suggest they've achieved some promissing preclinical results in mouse models, though the company hasn't confirmed this publically.",
-  "timeline": "- **2020-06-15** | Tempo incorporated in Delaware by [Quinten Lee](people/quinten-lee-24)\n- **2021-02-08** | Closed seed round, terms undisclosed\n- **2021-09-22** | [Yara Moore](people/yara-moore-174) formally joins as strategic advisor\n- **2022-04-11** | Published first platform paper in Journal of Computational Biology\n- **2022-11-30** | Moved into expanded South San Francisco lab facility\n- **2023-03-17** | Series A closed, reportedly oversubscribed\n- **2023-08-05** | Hired VP of Biology from Amgen\n- **2024-01-22** | Internal milestone: lead compound identified for T2D program\n- **2024-09-14** | Quinten Lee presented at JP Morgan Healthcare Conference (private session)\n- **2025-02-28** | Initiated IND-enabling studies for lead metabolic compound",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/tempo-24",
-    "name": "Tempo",
-    "category": "startup",
-    "industry": "biotech",
-    "founded_year": 2020,
-    "founders": [
-      "people/quinten-lee-24"
-    ],
-    "employees": [
-      "people/mia-liu-134"
-    ],
-    "advisors": [
-      "people/yara-moore-174"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__tessera-15.json b/eval/data/world-v1/companies__tessera-15.json
deleted file mode 100644
index fbe48dcb9..000000000
--- a/eval/data/world-v1/companies__tessera-15.json
+++ /dev/null
@@ -1,27 +0,0 @@
-{
-  "slug": "companies/tessera-15",
-  "type": "company",
-  "title": "Tessera",
-  "compiled_truth": "Tessera is a fintech startup founded in 2024 by [Noah Kapoor](people/noah-kapoor-15), a serial entrepreneur with deep expertise in payments infrastructure and distributed systems. The company is building what it describes as \"programmable treasury rails\" — essentially a platform that lets mid-market companies automate complex cash management workflows without relying on legacy banking integrations. Think of it as Plaid meets Airflow, but for corporate finance teams who are tired of moving money through spreadsheets and manual wire transfers.\n\nThe founding team is lean but credible. Noah previously spent six years at Stripe, where he led a team focused on cross-border settlement optimization. Before that, he did a stint at a Series B payments company that got aquired by Block in 2021. He's known in fintech circles for his pragmatic approach to product development — shipping fast, iterating based on real customer feedback, and avoiding the trap of over-engineering.\n\nTessera raised a $4.2M seed round in late 2024, led by [Kate Lopez](people/kate-lopez-99), a partner at Foundry Ventures who has backed several successful fintech exits. The round included participation from a handful of angel investors, mostly former operators from Stripe, Ramp, and Modern Treasury. The company has been relatively quiet about its traction, though Noah has mentioned in a few podcast appearances that they have \"a handful of design partners\" actively using the platform.\n\nOn the advisory side, [Yara Singh](people/yara-singh-195) has been helping the team think through go-to-market strategy and enterprise sales motions. Yara's background in scaling B2B fintech products has been valuable as Tessera figures out how to position itself against incumbents like Kyriba and newer players like Treasure.\n\nThe company operates out of San Francisco, with a small remote-first team of about eight people. Noah has been vocal about keeping the team small until they nail product-market fit — a lesson he says he learned the hard way at his previous startup. Tessera's current focus is on onboarding its first ten paying customers and proving out unit economics before raising a Series A, likely in late 2025.",
-  "timeline": "- **2024-02-12** | Noah Kapoor incorporates Tessera in Delaware, begins recruiting co-founding engineers\n- **2024-04-08** | First prototype of treasury automation platform demoed to potential design partners\n- **2024-06-15** | [Kate Lopez](people/kate-lopez-99) leads $4.2M seed round; Foundry Ventures announces the investment\n- **2024-07-22** | [Yara Singh](people/yara-singh-195) joins as formal advisor, focusing on GTM strategy\n- **2024-09-03** | Tessera onboards first two design partners — both mid-market e-commerce companies\n- **2024-11-18** | Noah speaks at Fintech Devcon about \"rethinking treasury infrastructure for the API era\"\n- **2025-01-09** | Team grows to eight; hires head of engineering from Modern Treasury\n- **2025-03-14** | Closes first paying customer contract, $48K ARR\n- **2025-05-02** | Begins early conversations with Series A investors, targeting Q4 2025 raise",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/tessera-15",
-    "name": "Tessera",
-    "category": "startup",
-    "industry": "fintech",
-    "founded_year": 2024,
-    "founders": [
-      "people/noah-kapoor-15"
-    ],
-    "investors": [
-      "people/kate-lopez-99"
-    ],
-    "employees": [
-      "people/gabe-wilson-125"
-    ],
-    "advisors": [
-      "people/yara-singh-195"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__umbra-48.json b/eval/data/world-v1/companies__umbra-48.json
deleted file mode 100644
index ec943ba86..000000000
--- a/eval/data/world-v1/companies__umbra-48.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "companies/umbra-48",
-  "type": "company",
-  "title": "Umbra",
-  "compiled_truth": "Umbra is a health tech startup founded in early 2025 by [Zoe Kim](people/zoe-kim-48), a serial entrepreneur with deep roots in digital therapeutics and wearable technology. The company operates in stealth mode for much of its first year, though insiders describe its focus as \"ambient health monitoring\" — a system that passively collects biometric and environmental data to surface early warning signs of chronic disease.\n\nThe founding thesis emerged from Kim's frustration with reactive healthcare models. She wanted to build something that could catch problems before they became crises, particulary for populations underserved by traditional primary care. Umbra's initial product combines low-power sensor hardware with an AI backend trained on longitudinal health data. The company has been tight-lipped about specifics, but demo videos leaked in mid-2025 showed a small wearable device syncing with ambient sensors placed around a home.\n\nAdvisory support comes from [Vera Rodriguez](people/vera-rodriguez-171), who brings credibility from her work in regulatory strategy and medical device commercialization. Rodriguez's involvement suggests Umbra is serious about FDA clearance and clinical validation, not just consumer wellness claims. Her network has reportedly helped the company secure early conversations with payer organizations interested in preventive care pilots.\n\nUmbra operates with a lean team — roughly twelve people as of late 2025, split between engineering, clinical research, and ops. They've raised a seed round, though the amount remains undisclosed. The company culture leans heavily on asynchronous work and documentation, a hallmark of Kim's previous ventures. Recruiting has focused on candidates with backgrounds in signal processing, embedded systems, and health informatics.\n\nThe competitive landscape is crowded, but Umbra differentiates itself by targeting B2B2C partnerships rather than direct-to-consumer sales. Early pilots with regional health systems are expected to begin in Q1 2026. Whether Umbra can execute on its ambitious vision remains to be seen, but the team's pedigree and early traction have attracted attention from health-focused VCs watching the space closely.",
-  "timeline": "- **2025-01-18** | Umbra incorporated in Delaware by [Zoe Kim](people/zoe-kim-48)\n- **2025-02-04** | [Vera Rodriguez](people/vera-rodriguez-171) joins as lead advisor\n- **2025-03-22** | Seed round closed, amount undisclosed\n- **2025-04-10** | First engineering hire — embedded systems lead from Oura\n- **2025-06-15** | Internal prototype v0.1 completed; early testing begins\n- **2025-08-07** | Demo video leaked on Twitter, sparking industry speculation\n- **2025-09-30** | Team reaches 12 full-time employees\n- **2025-11-12** | Preliminary conversations with two regional health systems for pilot programs\n- **2026-01-08** | Planned kickoff for first B2B2C pilot deployment",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/umbra-48",
-    "name": "Umbra",
-    "category": "startup",
-    "industry": "health tech",
-    "founded_year": 2025,
-    "founders": [
-      "people/zoe-kim-48"
-    ],
-    "employees": [
-      "people/julia-garcia-158"
-    ],
-    "advisors": [
-      "people/vera-rodriguez-171"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__vector-6.json b/eval/data/world-v1/companies__vector-6.json
deleted file mode 100644
index f89a5daec..000000000
--- a/eval/data/world-v1/companies__vector-6.json
+++ /dev/null
@@ -1,30 +0,0 @@
-{
-  "slug": "companies/vector-6",
-  "type": "company",
-  "title": "Vector",
-  "compiled_truth": "Vector is a health tech startup founded in 2020 by [Uma Brown](people/uma-brown-6), focused on building AI-powered diagnostic tools for early disease detection. The company emerged from Uma's frustration with the slow pace of traditional diagnostic workflows in clinical settings. Based in Austin, Texas, Vector has positioned itself at the intersection of machine learning and medical imaging, with their flagship product analyzing radiology scans to flag potential anomalies before they become critical.\n\nThe startup has attracted notable backing from angel investors including [David Zhang](people/david-zhang-83), [Vera Gonzalez](people/vera-gonzalez-103), and [Grace Martinez](people/grace-martinez-109). Zhang in particular has been instrumental in connecting Vector with enterprise healthcare networks through his existing portfolio companies. The advisory board includes [Wendy Wilson](people/wendy-wilson-170) and [Bob Chen](people/bob-chen-185), both of whom bring deep experiance in healthcare compliance and regulatory strategy—critical for a company operating in such a heavily regulated space.\n\nVector's approach differs from competitors by focusing on integration rather than replacement. Their software plugs into existing hospital PACS systems, meaning radiologists don't need to change their workflows dramatically. This pragmatic stance has helped them secure pilot programs with three regional hospital networks, though they haven't disclosed names publicly. The company claims a 23% improvement in early detection rates for certain cancers based on internal studies, though peer-reviewed validation is still pending.\n\nUma Brown serves as CEO and has been the public face of the company at industry conferences. She's known for being blunt about the limitations of AI in healthcare, which has ironically helped build trust with skeptical clinicians. Recent moves include expanding the engineering team from 8 to 15 people and opening a small office in Boston to be closer to major academic medical centers.\n\nVector remains a seed-stage company but is reportedly preparing for a Series A round in late 2024. The health tech space is crowded, but their focus on practical integration and Uma's credibility in the space gives them a fighting chance. Challenges remain around FDA clearance timelines and convincing risk-averse hospital administrators to adopt new technology.",
-  "timeline": "- **2020-03-15** | Vector incorporated in Delaware by [Uma Brown](people/uma-brown-6)\n- **2020-09-02** | Closed pre-seed round with [David Zhang](people/david-zhang-83) and [Vera Gonzalez](people/vera-gonzalez-103) participating\n- **2021-04-18** | First prototype deployed for internal testing with synthetic medical data\n- **2021-11-30** | [Wendy Wilson](people/wendy-wilson-170) joins advisory board to help navigate FDA pathway\n- **2022-06-14** | Signed first hospital pilot agreement (name under NDA)\n- **2023-02-22** | [Grace Martinez](people/grace-martinez-109) invests in bridge round; joins cap table\n- **2023-08-09** | Uma Brown presents early detection results at HealthTech Summit Austin\n- **2024-01-17** | Boston office opened to strengthen academic medical center relationships\n- **2024-05-03** | Engineering team expansion completed, now at 15 full-time employees\n- **2025-02-11** | FDA pre-submission meeting scheduled for Q2 2025",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/vector-6",
-    "name": "Vector",
-    "category": "startup",
-    "industry": "health tech",
-    "founded_year": 2020,
-    "founders": [
-      "people/uma-brown-6"
-    ],
-    "investors": [
-      "people/david-zhang-83",
-      "people/vera-gonzalez-103",
-      "people/grace-martinez-109"
-    ],
-    "employees": [
-      "people/victor-jackson-116"
-    ],
-    "advisors": [
-      "people/wendy-wilson-170",
-      "people/bob-chen-185"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__vector-labs-56.json b/eval/data/world-v1/companies__vector-labs-56.json
deleted file mode 100644
index b33e275be..000000000
--- a/eval/data/world-v1/companies__vector-labs-56.json
+++ /dev/null
@@ -1,28 +0,0 @@
-{
-  "slug": "companies/vector-labs-56",
-  "type": "company",
-  "title": "Vector Labs",
-  "compiled_truth": "Vector Labs is a robotics startup founded in early 2025 by [Yara Kim](people/yara-kim-56), a mechanical engineer with a background in autonomous systems. The company emerged from Kim's frustration with the fragmented state of warehouse automation—too many point solutions, not enough integration. Vector's core product is a modular robotic platform designed for mid-sized logistics operations, the kind of facilities that cant afford the massive capital outlay of a fully automated Amazon-style warehouse but still need to scale beyond manual labor.\n\nThe company operates out of a converted industrial space in Oakland, California, where a small team of engineers iterates rapidly on hardware prototypes. Their approach is somewhat unconventional: rather than building robots from scratch, Vector Labs focuses on retrofit kits that can upgrade existing conveyor systems and pallet movers with autonomous capabilities. This strategy keeps costs down and shortens deployment timelines, which has resonated with early pilot customers.\n\nFunding came through a seed round led by [Iris Lee](people/iris-lee-82), a prolific angel investor known for backing deep-tech companies. Lee reportedly wrote the first check after seeing a demo at a hardware meetup in San Francisco. The round also included a handful of other angels, though Vector hasn't disclosed the full list or total amount raised.\n\nOn the advisory side, Vector Labs has assembled a small but experienced board. [Yara Moore](people/yara-moore-174) brings operational expertise from her years scaling manufacturing startups, while [Yara Singh](people/yara-singh-195) contributes technical depth in computer vision and sensor fusion. Both advisors are hands-on, attending weekly syncs and occasionally visiting the Oakland lab to review progress.\n\nVector Labs is still pre-revenue as of mid-2025, though the team claims to have letters of intent from three regional distribution companies. The robotics space is crowded and capital-intensive, but Vector's lean approach and focus on retrofitting could carve out a defensible niche. Kim has been vocal about avoiding the trap of over-engineering—ship fast, learn faster. Whether that philosophy scales remains to be seen.",
-  "timeline": "- **2025-01-14** | Vector Labs incorporated in Delaware by founder Yara Kim\n- **2025-02-03** | Closed seed round with [Iris Lee](people/iris-lee-82) as lead investor\n- **2025-02-20** | Signed lease on Oakland warehouse space for R&D operations\n- **2025-03-08** | [Yara Moore](people/yara-moore-174) joined as official advisor\n- **2025-03-22** | First functional prototype of retrofit automation kit completed\n- **2025-04-10** | [Yara Singh](people/yara-singh-195) began advising on sensor integration\n- **2025-05-15** | Began pilot deployment discussions with regional logistics company\n- **2025-06-02** | Hired third full-time engineer, expanding core team to five\n- **2025-06-19** | Yara Kim presented at Bay Area Hardware Founders meetup",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/vector-labs-56",
-    "name": "Vector Labs",
-    "category": "startup",
-    "industry": "robotics",
-    "founded_year": 2025,
-    "founders": [
-      "people/yara-kim-56"
-    ],
-    "investors": [
-      "people/iris-lee-82"
-    ],
-    "employees": [
-      "people/owen-smith-166"
-    ],
-    "advisors": [
-      "people/yara-moore-174",
-      "people/yara-singh-195"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__vellum-49.json b/eval/data/world-v1/companies__vellum-49.json
deleted file mode 100644
index c9d157c56..000000000
--- a/eval/data/world-v1/companies__vellum-49.json
+++ /dev/null
@@ -1,27 +0,0 @@
-{
-  "slug": "companies/vellum-49",
-  "type": "company",
-  "title": "Vellum",
-  "compiled_truth": "Vellum is an AI applications startup founded in 2021 by [Chris Davis](people/chris-davis-49), a serial entrepreneur with deep roots in machine learning infrastructure. The company positions itself as a development platform for building production-grade LLM applications, offering tools for prompt engineering, model evaluation, and workflow orchestration. Since its inception, Vellum has attracted attention from investors who see potential in the picks-and-shovels approach to the generative AI boom.\n\nThe platform's core value proposition centers on helping engineering teams move from prototype to production without the typical headaches. Vellum provides version control for prompts, A/B testing capabilities, and monitoring dashboards that track model performance over time. It's a pragmatic play—rather than building foundation models, they're building the tooling layer that makes those models actually useable in enterprise contexts.\n\nFunding for the company came from [Rosa Miller](people/rosa-miller-98), who led an early seed round that helped Vellum scale its engineering team from three to fifteen people within eighteen months. Rosa saw in Chris's vision something that resonated with her thesis on infrastructure plays during platform shifts. The bet seems to be paying off as Vellum has landed several mid-market customers in fintech and healthtech verticals.\n\nOn the advisory side, [Rachel Gonzalez](people/rachel-gonzalez-175) has been instrumental in shaping go-to-market strategy. Her background in enterprise sales helped the company refine its pricing model and identify the right buyer personas within target organizations. Rachel pushed the team to focus on developer experience as a differentiator, arguing that bottoms-up adoption would be critical in a crowded market.\n\nVellum faces stiff competiton from both well-funded startups and the model providers themselves, who are increasingly bundling similar capabilities. But Chris Davis remains confident that specialization wins. The company's recent pivot toward supporting multi-model workflows—allowing customers to route between different LLMs based on cost, latency, or capability—has opened up new use cases. As of mid-2024, Vellum processes over 50 million API calls monthly, a figure that's grown 4x year-over-year.",
-  "timeline": "- **2021-03-15** | Vellum incorporated in Delaware by [Chris Davis](people/chris-davis-49)\n- **2021-09-02** | Closed $2.1M seed round led by [Rosa Miller](people/rosa-miller-98)\n- **2022-04-18** | Launched public beta of prompt management platform\n- **2022-11-07** | [Rachel Gonzalez](people/rachel-gonzalez-175) joins as strategic advisor\n- **2023-02-22** | Announced SOC 2 Type II compliance certification\n- **2023-08-14** | Shipped multi-model routing feature, dubbed \"Model Router\"\n- **2024-01-09** | Chris Davis keynoted at AI Infrastructure Summit in San Francisco\n- **2024-06-30** | Reached 50M monthly API calls milestone\n- **2025-03-11** | Opened first international office in London\n- **2025-09-05** | Partnership announced with major cloud provider for native integration",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/vellum-49",
-    "name": "Vellum",
-    "category": "startup",
-    "industry": "AI applications",
-    "founded_year": 2021,
-    "founders": [
-      "people/chris-davis-49"
-    ],
-    "investors": [
-      "people/rosa-miller-98"
-    ],
-    "employees": [
-      "people/noah-chen-159"
-    ],
-    "advisors": [
-      "people/rachel-gonzalez-175"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__vox-25.json b/eval/data/world-v1/companies__vox-25.json
deleted file mode 100644
index 1a3c1e932..000000000
--- a/eval/data/world-v1/companies__vox-25.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "companies/vox-25",
-  "type": "company",
-  "title": "Vox - AI Applications Startup",
-  "compiled_truth": "Vox is an early-stage AI applications company founded in 2023 by [Vera Wilson](people/vera-wilson-25), a serial entrepreneur with a background in natural language processing and enterprise software. The startup focuses on building conversational AI tools for customer service automation, targeting mid-market SaaS companies that need scalable support solutions without the overhead of large teams.\n\nThe company emerged from Vera's frustration with existing chatbot solutions, which she found clunky and unable to handle nuanced customer inquiries. Vox's core product uses a proprietary fine-tuning approach that allows businesses to train AI agents on their specific knowledge bases in under 24 hours. Early adopters have reported significant reductions in ticket volume, though the company hasn't published official metrics yet.\n\nFunding came through an angel round led by [Fiona Moore](people/fiona-moore-88), who had previously backed two of Vera's colleagues from her time at a larger tech firm. Fiona's involvement brought not just capital but also connections to potential enterprise customers in the fintech space. The exact amount raised hasn't been disclosed publicly, though sources suggest it was in the low seven figures.\n\nVox operates with a lean team of about eight people, mostly engineers, working out of a shared office space in San Francisco's SoMa district. The company has been relatively quiet in terms of public presence, preferring to focus on product developement over marketing. Wilson has said in interviews that she wants to \"let the product speak for itself\" before ramping up sales efforts.\n\nThe competitive landscape is crowded, with players like Intercom and newer AI-native startups vying for the same customers. Vox differentiates itself through speed of deployment and pricing—offering a usage-based model rather than expensive annual contracts. Whether this strategy will hold up as they scale remains to be seen, but early traction suggests theres genuine demand for what they're building.",
-  "timeline": "- **2023-03-15** | Vox incorporated in Delaware by [Vera Wilson](people/vera-wilson-25)\n- **2023-05-22** | Closed angel round with [Fiona Moore](people/fiona-moore-88) as lead investor\n- **2023-07-10** | First engineering hire joins from Google's Bard team\n- **2023-09-04** | Alpha version of conversational AI platform launched to 5 pilot customers\n- **2024-01-18** | Expanded to 12 paying customers, mostly fintech startups\n- **2024-04-30** | Vera Wilson spoke at AI Summit SF about rapid deployment strategies\n- **2024-08-12** | Moved into new office space in SoMa, team grew to 8 employees\n- **2024-11-05** | Launched self-serve onboarding for smaller customers\n- **2025-02-20** | Partnership announced with a mid-sized CRM vendor for native integration",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/vox-25",
-    "name": "Vox",
-    "category": "startup",
-    "industry": "AI applications",
-    "founded_year": 2023,
-    "founders": [
-      "people/vera-wilson-25"
-    ],
-    "investors": [
-      "people/fiona-moore-88"
-    ],
-    "employees": [
-      "people/yara-zhang-135"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__wisp-26.json b/eval/data/world-v1/companies__wisp-26.json
deleted file mode 100644
index 6cff02b95..000000000
--- a/eval/data/world-v1/companies__wisp-26.json
+++ /dev/null
@@ -1,21 +0,0 @@
-{
-  "slug": "companies/wisp-26",
-  "type": "company",
-  "title": "Wisp",
-  "compiled_truth": "Wisp is an edtech startup founded in 2023 by [Linda Kim](people/linda-kim-26), a former learning sciences researcher who spent years studying how students retain information. The company's core product is an AI-powered microlearning platform that delivers personalized knowledge snippets throughout the day, designed to fit into the gaps between meetings, commutes, and coffee breaks.\n\nThe premise behind Wisp is deceptively simple: most people don't have time for dedicated learning sessions, but they do have dozens of small moments scattered across their day. Wisp's algorythm identifies these windows and pushes bite-sized lessons that adapt based on user engagement and retention patterns. The platform currently focuses on professional development content—soft skills, technical upskilling, and industry-specific knowledge modules.\n\n[Linda Kim](people/linda-kim-26) has been vocal about her frustration with traditional corporate training, calling most of it \"expensive theater that nobody remembers.\" She built the initial prototype while still finishing her PhD, testing it with a small cohort of graduate students before pivoting to enterprise customers. Her academic background gives Wisp some credibility in a crowded market full of flashy apps with questionable pedagogical foundations.\n\nThe startup raised a small pre-seed round in late 2023, though exact figures haven't been disclosed. They've been operating lean, with a team of about eight people split between engineering and content development. Early traction has come primarily from mid-sized tech companies looking for alternatives to clunky LMS platforms that employees actively avoid.\n\nWisp faces significant competition from established players like Coursera for Business and newer entrants in the microlearning space. What differentiates them, according to Kim, is their focus on spaced repetition and contextual delivery rather than gamification gimmicks. The company has been experimenting with intergrations into Slack and Microsoft Teams, meeting users where they already work rather than asking them to open another app.\n\nRecent moves suggest Wisp is preparing for a seed raise sometime in 2024, with plans to expand their content library and build out more robust analytics for L&D teams.",
-  "timeline": "- **2023-02-14** | [Linda Kim](people/linda-kim-26) incorporates Wisp after testing early prototype with graduate students\n- **2023-05-08** | First version of the Wisp mobile app launches in private beta with 200 users\n- **2023-07-22** | Closes undisclosed pre-seed round from angel investors\n- **2023-09-15** | Hires first full-time engineer, expanding team to four people\n- **2023-11-03** | Lands first enterprise pilot with a Series B fintech company\n- **2024-01-19** | Launches Slack integration for seamless in-workflow learning delivery\n- **2024-04-11** | [Linda Kim](people/linda-kim-26) speaks at EdTech Week about the future of workplace learning\n- **2024-08-26** | Reaches 15,000 active daily users across enterprise accounts\n- **2025-02-07** | Begins seed fundraising process, targeting $3M round",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/wisp-26",
-    "name": "Wisp",
-    "category": "startup",
-    "industry": "edtech",
-    "founded_year": 2023,
-    "founders": [
-      "people/linda-kim-26"
-    ],
-    "employees": [
-      "people/david-liu-136"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/companies__zenith-27.json b/eval/data/world-v1/companies__zenith-27.json
deleted file mode 100644
index 718f65866..000000000
--- a/eval/data/world-v1/companies__zenith-27.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "companies/zenith-27",
-  "type": "company",
-  "title": "Zenith",
-  "compiled_truth": "Zenith is a logistics startup founded in 2018 by [Priya Zhang](people/priya-zhang-27), who saw an opportunity to modernize last-mile delivery infrastructure for mid-sized e-commerce brands. The company operates primarily in the B2B space, offering a software platform that optimizes routing, warehouse allocation, and carrier selection for businesses that have outgrown basic shipping solutions but aren't large enough to build proprietary systems.\n\nThe founding story is fairly straightforward. Zhang had spent several years in supply chain consulting and kept running into the same problem: companies doing $5-50M in annual revenue were stuck between consumer-grade tools and enterprise solutions they couldn't afford. Zenith launched with a simple routing optimizer and has since expanded into a fuller suite including inventory forecasting and returns managment.\n\nGrowth has been steady if not spectacular. The company raised a seed round in 2019 and a Series A in late 2021, though exact figures haven't been publicly disclosed. They've been somewhat quiet compared to flashier logistics tech players, preferring to focus on unit economics over growth-at-all-costs. This measured approach has attracted interest from experienced operators in the space.\n\n[Tina Moore](people/tina-moore-191) joined as an advisor sometime in 2022, bringing her background in operations scaling to help Zenith think through their expansion strategy. Her involvement signaled a shift toward more aggressive market positioning, though the company has remained disciplined about customer acquisition costs.\n\nThe team is currently around 45 people, split between engineering in San Francisco and a customer success hub in Austin. They've been experimenting with AI-driven demand prediction features, which Zhang has mentioned in a few podcast appearances as a major focus for the next 18 months. Some early customers have reported 15-20% reductions in shipping costs after implementing the platform, though these numbers are self-reported and should be taken with apropriate skepticism.\n\nZenith competes with larger players like Shippo and ShipBob but differentiates on flexibility and pricing for the mid-market segment. The logistics tech space remains crowded, and it's unclear whether Zenith can carve out a durable niche or will eventually need to consolidate.",
-  "timeline": "- **2018-03-12** | Zenith incorporated in Delaware; [Priya Zhang](people/priya-zhang-27) begins building initial MVP\n- **2019-06-08** | Closed seed round from logistics-focused angels; first three pilot customers onboarded\n- **2020-11-15** | Platform processed 1 millionth shipment; expanded routing to cover all 50 states\n- **2021-09-22** | Series A closed; company moves to larger SF office space\n- **2022-04-03** | [Tina Moore](people/tina-moore-191) joins advisory board to help with operational scaling\n- **2023-01-17** | Launched returns management module after six months of beta testing\n- **2023-08-29** | Zhang speaks at LogiTech Summit on mid-market logistics challenges\n- **2024-02-14** | Austin customer success hub opens with initial team of 12\n- **2024-10-06** | Beta release of AI demand forecasting feature to select customers\n- **2025-03-21** | Reached 200 active enterprise customers milestone",
-  "_facts": {
-    "type": "company",
-    "slug": "companies/zenith-27",
-    "name": "Zenith",
-    "category": "startup",
-    "industry": "logistics",
-    "founded_year": 2018,
-    "founders": [
-      "people/priya-zhang-27"
-    ],
-    "employees": [
-      "people/eve-nakamura-137"
-    ],
-    "advisors": [
-      "people/tina-moore-191"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__agentic-workflows.json b/eval/data/world-v1/concepts__agentic-workflows.json
deleted file mode 100644
index fae6f0d4c..000000000
--- a/eval/data/world-v1/concepts__agentic-workflows.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "concepts/agentic-workflows",
-  "type": "concept",
-  "title": "Agentic Workflows",
-  "compiled_truth": "Agentic workflows represent a paradigm shift in how we think about software systems and, by extension, how we think about building companies around AI. The core idea is deceptively simple: instead of monolithic models handling tasks end-to-end, you orchestrate multiple specialized agents that collaborate, delegate, and iterate. Each agent has a defined role, access to specific tools, and the autonomy to make decisions within its domain.\n\nThis isn't just a technical architecture—it's a strategic frame. When we evaluate companies through the agentic lens, we're asking: does this team understand that the future of work involves humans and AI agents operating as peers in complex workflows? Are they building for a world where agents hire other agents, where supervision becomes the primary human function?\n\n[Brink Labs](companies/brink-labs-79) exemplifies this thinking in their approach to developer tooling. Rather than building another copilot that suggests code, they've constructed an ecosystem where coding agents can spawn sub-agents for testing, documentation, and deployment. The human developer becomes an architect and reviewer, not a typist. It's a subtle but crucial distinction that many competitors miss entirely.\n\nThe implications for company building are profound. Traditional SaaS metrics don't capture value creation when your product enables agentic orchestration. Seat-based pricing breaks down. Usage patterns become non-linear and harder to predict. [Mosaic Labs](companies/mosaic-labs-64) has been wrestling with this directly—their pricing model evolved three times in 2024 alone as they discovered that enterprise customers wanted to pay for outcomes, not API calls.\n\nWe're also seeing agentic workflows reshape org design itself. [Quantum](companies/quantum-7) runs what they call \"agent-first operations\" where internal processes are designed assuming AI agents will execute most steps. Humans define goals, set constraints, and handle exceptions. It sounds dystopian until you realize it frees people to do genuinely creative work.\n\nThe risks are real though. Agentic systems can fail in cascading, unpredictable ways. Debugging becomes archaeology. And there's a talent gap—few engineers have intuition for designing robust multi-agent systems. The companes that win will be those who treat agentic thinking as a core competency, not a feature checkbox.",
-  "timeline": "- **2021-08-14** | First internal memo on \"agent orchestration\" as investment thesis, inspired by early AutoGPT experiments\n- **2022-03-22** | Hosted dinner with founders exploring multi-agent architectures; 12 attendees including early [Brink Labs](companies/brink-labs-79) team\n- **2023-01-09** | Published blog post \"Beyond Copilots\" arguing for agentic workflows as the next platform shift\n- **2023-06-17** | [Mosaic Labs](companies/mosaic-labs-64) seed investment; thesis centered on agentic workflows for enterprise ops\n- **2023-11-03** | Attended AgentCon in SF; noted shift from academic curiosity to production deployments\n- **2024-02-28** | Internal workshop with portfolio companies on agent supervision patterns and failure modes\n- **2024-07-11** | [Quantum](companies/quantum-7) presents agent-first operations model at LP meeting; strong reception\n- **2024-10-19** | Began tracking \"agentic readiness\" as evaluation criteria for new deals\n- **2025-01-06** | Partnered with Stanford HAI on research into human-agent collaboration frameworks\n- **2025-04-22** | Updated thesis to emphasize agent-to-agent commerce as emerging pattern",
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/agentic-workflows",
-    "name": "agentic workflows",
-    "description": "agentic workflows as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/brink-labs-79",
-      "companies/mosaic-labs-64",
-      "companies/quantum-7"
-    ],
-    "related_people": [
-      "people/quinten-wang-17",
-      "people/nina-rodriguez-18"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__ai-first-product.json b/eval/data/world-v1/concepts__ai-first-product.json
deleted file mode 100644
index ebae9d7a7..000000000
--- a/eval/data/world-v1/concepts__ai-first-product.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "concepts/ai-first-product",
-  "type": "concept",
-  "title": "AI-First Product",
-  "compiled_truth": "AI-first product is a strategic frame for building companies where artificial intelligence isn't bolted on as a feature but serves as the foundational architecture from day one. Unlike traditional software that treats ML as an optimization layer, AI-first products are designed around the assumption that intelligence is the core value proposition. The user experience, data flywheel, and business model all flow from this premise.\n\nThe distinction matters more than most founders realize. A traditional SaaS tool with AI features still operates on deterministic logic—the AI enhances but doesn't define. An AI-first product inverts this: the model's capabilities shape what's even possible to build. [Wisp Labs](companies/wisp-labs-76) exemplifies this approach in their agent infrastructure work, where the entire product surface emerges from what autonomous systems can reliably accomplish.\n\nThere's a tension here worth naming. AI-first doesn't mean AI-only. The best implementations pair model capabilities with thoughtful UX constraints that guide users toward productive outcomes. [Apex](companies/apex-18) has been particularly clever about this—their interface feels magical precisely because they've hidden the probabilistic nature of the underlying system behind carefully designed interaction patterns.\n\nFrom an investment persepctive, AI-first companies exhibit different scaling dynamics. Traditional SaaS scales with seat count; AI-first products often scale with usage intensity or data volume. This creates interesting unit economics that don't map cleanly to established benchmarks. Gross margins can look worse initially due to inference costs, but the moat potential is substantially higher when the product improves with every interaction.\n\n[Cascade](companies/cascade-30) represents another variant—they've built AI-first infrastructure that other companies use to become AI-first themselves. It's a picks-and-shovels play on the broader trend. Their recent traction suggests the market is maturing past the experimentation phase.\n\nThe frame also implies organizational choices. AI-first companies need research eng capabilities earlier than typical startups. They need to think about evaluation and testing differently. They ship with more uncertainty about edge cases. The best teams embrace this ambiguity rather than fighting it, treating their products as living systems that evolve alongside frontier model improvements.",
-  "timeline": "- **2021-08-14** | First internal memo outlining AI-first as distinct investment thesis, contrasting with AI-enabled\n- **2022-03-22** | Partnered with [Wisp Labs](companies/wisp-labs-76) as thesis validation—their agent-native approach matched the framework\n- **2022-11-09** | Published thinking on AI-first unit economics; unexpected traction on Twitter sparked several inbound deals\n- **2023-04-17** | [Cascade](companies/cascade-30) seed investment; infrastructure layer for AI-first development\n- **2023-09-03** | Internal debate on whether AI-first framing was becoming too broad, losing analytical value\n- **2024-02-28** | [Apex](companies/apex-18) Series A co-lead; strongest portfolio example of AI-first UX principles\n- **2024-07-11** | LP meeting presentation on AI-first portfolio performance vs traditional SaaS cohorts\n- **2025-01-19** | Revised thesis to account for inference cost deflation and its impact on margin profiles\n- **2025-05-02** | Hosted dinner with 12 AI-first founders to discuss emerging patterns in go-to-market",
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/ai-first-product",
-    "name": "AI-first product",
-    "description": "AI-first product as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/wisp-labs-76",
-      "companies/apex-18",
-      "companies/cascade-30"
-    ],
-    "related_people": [
-      "people/rachel-park-64",
-      "people/alice-kim-41"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__carbon-credits.json b/eval/data/world-v1/concepts__carbon-credits.json
deleted file mode 100644
index 4d0084628..000000000
--- a/eval/data/world-v1/concepts__carbon-credits.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "concepts/carbon-credits",
-  "type": "concept",
-  "title": "Carbon Credits",
-  "compiled_truth": "Carbon credits as a strategic frame for company building represents a mental model borrowed from environmental markets but applied to organizational decision-making. The core thesis: every company accumulates both positive and negative capital across multiple dimensions—technical debt, cultural equity, market positioning, team morale—and these can be traded against each other in ways that mirror carbon offset markets.\n\nThe framework emerged from conversations with founders at [Talon Labs](companies/talon-labs-97) who noticed that their aggressive shipping velocity was creating what they called 'cultural debt.' They were burning through goodwill with their engineering team by pushing unsustainable timelines, but the market wins were generating enough momentum to 'offset' this through equity appreciation and hiring leverage. The question became: how do you account for these trades explicitly rather than stumbling into them?\n\n[Lucid Labs](companies/lucid-labs-71) took this further by actually tracking what they call 'credit balances' across five dimensions: technical, cultural, financial, reputational, and operational. Their thesis is that most startups fail not because they run out of money, but because they overdraw on one of these accounts without realizing it. You can ship fast and accumulate technical debt, but only if you're simultaneously building cultural credits through transparency about the tradeoffs.\n\nThe carbon credits model suggests several non-obvious strategies. First, credits are not fungible—you can't always trade financial capital for cultural capital, especially once trust is broken. Second, some credits compound while others decay. Market positioning credits tend to compound; operational credits (like documentation) decay if not maintained. Third, the exchange rates between credit types shift based on company stage. Early on, technical debt is cheap becuase you might pivot anyway. Later, that same debt becomes ruinous.\n\n[Sentinel](companies/sentinel-23) reportedly uses a quarterly 'credit audit' in their planning process, explicitly naming which accounts they're drawing down and which they're depositing into. This framing helps avoid the common failure mode where founders optimize for one metric while unknowingly bankrupting another. The carbon credits lens doesn't prescribe what to optimize for—it just makes the trades visible.",
-  "timeline": "- **2021-08-14** | Initial framework sketched during founder dinner with [Talon Labs](companies/talon-labs-97) team, discussing technical debt tradeoffs\n- **2022-02-03** | [Lucid Labs](companies/lucid-labs-71) adopts five-dimension credit tracking in their planning docs\n- **2022-09-19** | Framework shared in private founder Slack, sparks heated debate about quantifying culture\n- **2023-01-27** | [Sentinel](companies/sentinel-23) implements quarterly credit audits, first company to formalize the process\n- **2023-06-11** | Carbon credits concept mentioned in podcast interview, gains wider circulation\n- **2023-11-04** | Counter-arguments emerge: critics say framework encourages transactional thinking about team dynamics\n- **2024-03-22** | [Talon Labs](companies/talon-labs-97) shares internal retro showing they overdrew cultural credits in 2023, lessons learned\n- **2024-08-30** | Framework adapted for investor communications—founders using it to explain strategic tradeoffs in board decks\n- **2025-02-15** | Several seed-stage companies now reference carbon credits in their operating docs",
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/carbon-credits",
-    "name": "carbon credits",
-    "description": "carbon credits as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/talon-labs-97",
-      "companies/lucid-labs-71",
-      "companies/sentinel-23"
-    ],
-    "related_people": [
-      "people/ian-kapoor-162",
-      "people/linda-taylor-178"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__churn-cohorts.json b/eval/data/world-v1/concepts__churn-cohorts.json
deleted file mode 100644
index cd0d430f4..000000000
--- a/eval/data/world-v1/concepts__churn-cohorts.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "concepts/churn-cohorts",
-  "type": "concept",
-  "title": "Churn Cohorts",
-  "compiled_truth": "Churn cohorts represent a strategic framework for understanding how different customer segments behave over time, particularly in SaaS and subscription businesses. The core insight is deceptively simple: not all churn is created equal. Customers acquired through different channels, at different price points, or during different market conditions will exhibit fundamentally different retention curves. Treating churn as a monolithic metric obscures the signal you actually need.\n\nThe framework becomes particularly powerful when applied to company building decisions. [Lattice](companies/lattice-39) demonstrated this well during their expansion from performance management into compensation—they discovered that customers who adopted multiple products in their first year had dramatically lower churn than single-product customers, even when controlling for company size. This wasn't just a correlation to track; it became a strategic imperative that reshaped their entire go-to-market motion.\n\nThinking in churn cohorts forces founders to ask harder questions. Which customer segments are actually profitable on a lifetime basis? Where is growth masking underlying retention problems? The temptation in hypergrowth is to celebrate net revenue retention while ignoring that your best cohorts are subsidizing terrible ones. [Foundry](companies/foundry-33) ran into this exact problem—their enterprise cohorts looked phenomenal while SMB churn was quietly destroying unit economics. Once they segmented properly, the decision to move upmarket became obvious.\n\nThere's also a temporal dimension worth considering. Cohorts acquired during economic expansions often churn harder during contractions. [Gravity](companies/gravity-17) built their forecasting models around this insight, helping portfolio companies stress-test retention assumptions against macroeconomic scenarios. The companies that survived 2022-2023 best were often those who had already identified their most resilient cohorts and doubled down.\n\nThe framework isn't without limitations. Over-segmentation can lead to analysis paralysis—you need enough data density in each cohort to draw meaningful conclusions. And cohort analysis is inherently backward-looking; it tells you what happened, not necesarily what will happen as you enter new markets or launch new products. Still, for any subscription business past initial product-market fit, churn cohort analysis should be table stakes.",
-  "timeline": "- **2021-03-15** | First internal memo on cohort-based retention analysis circulated among portfolio companies\n- **2021-09-22** | [Lattice](companies/lattice-39) presents multi-product cohort findings at offsite, sparks broader framework development\n- **2022-04-10** | Framework formally named \"churn cohorts\" in partner meeting notes\n- **2022-11-08** | [Foundry](companies/foundry-33) case study completed showing SMB vs enterprise retention divergence\n- **2023-02-14** | Workshop held with 12 portfolio companies on implementing cohort tracking in their data stacks\n- **2023-07-20** | [Gravity](companies/gravity-17) integrates churn cohort modeling into their scenario planning tools\n- **2024-01-30** | Published internal guide: \"Churn Cohorts: A Practical Framework for Subscription Businesses\"\n- **2024-09-12** | Concept referenced in board prep materials for 8 active investments\n- **2025-03-05** | Updated framework to include PLG-specific cohort considerations based on recent learnings",
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/churn-cohorts",
-    "name": "churn cohorts",
-    "description": "churn cohorts as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/lattice-39",
-      "companies/foundry-33",
-      "companies/gravity-17"
-    ],
-    "related_people": [
-      "people/xavier-patel-183",
-      "people/linda-taylor-178"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__community-led-growth.json b/eval/data/world-v1/concepts__community-led-growth.json
deleted file mode 100644
index b5017d738..000000000
--- a/eval/data/world-v1/concepts__community-led-growth.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "concepts/community-led-growth",
-  "type": "concept",
-  "title": "Community-Led Growth",
-  "compiled_truth": "Community-led growth represents a fundamental shift in how companies think about customer acquisition and retention. Rather than treating community as a marketing channel or support cost center, CLG positions community as the primary engine of company growth. The thesis is simple: when you build genuine spaces where users help each other, create content, and develop emotional investment in your product, you unlock compounding returns that paid acquisition can never match.\n\nThe model works best for products with natural network effects or where expertise sharing creates value. [Echo](companies/echo-32) exemplifies this approach—they've built their entire go-to-market around developer communities rather than traditional enterprise sales. Their community forums generate more qualified leads than their paid campaigns ever did, and at a fraction of the cost. The key insight from Echo's playbook: community members who help others become deeply invested in the product's success.\n\nWhat distinguishes CLG from traditional community marketing is the strategic primacy it gives to community health metrics. Companies like [Ranger Labs](companies/ranger-labs-72) track community engagement with the same rigor they apply to revenue. Active contributors, response times, content velocity—these become leading indicators for growth. Ranger's approach has been particularly effective in their expansion into new verticals; they seed communities with power users before launching sales motions.\n\nThere's a common misconception that community-led growth means abandoning sales teams. The reality is more nuanced. [Kindle](companies/kindle-20) runs a hybrid model where community surfaces intent signals that sales then acts on. Their community managers work closely with account executives, creating a feedback loop thats genuinely novel. This integrated approach has shortened their sales cycles considerably.\n\nThe challenges with CLG are real though. It requires patience—community compounding takes 12-18 months to show material impact. It demands authentic investment; users can smell performative community building instantly. And it's hard to attribute revenue cleanly, which makes CFOs nervous. But for companies willing to play the long game, community-led growth offers defensibility that traditional GTM motions simply cannot provide.",
-  "timeline": "- **2021-03-15** | First formal articulation of CLG framework in a16z blog post, sparking wider industry conversation\n- **2022-01-22** | [Echo](companies/echo-32) publicly credits community strategy for reaching $10M ARR without dedicated sales team\n- **2022-08-09** | Community-Led Growth Summit launches in San Francisco, 400+ attendees\n- **2023-02-14** | [Ranger Labs](companies/ranger-labs-72) publishes internal community metrics framework, becomes industry template\n- **2023-07-30** | Major critique published arguing CLG doesn't scale for enterprise—sparks healthy debate\n- **2023-11-18** | [Kindle](companies/kindle-20) case study on hybrid community-sales model presented at SaaStr\n- **2024-04-02** | CLG Slack community passes 8,000 members, becomes primary watering hole for practitioners\n- **2024-09-11** | First academic paper studying CLG ROI published by Stanford GSB researchers\n- **2025-01-27** | Growing consensus that CLG works best as complement to, not replacement for, traditional GTM",
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/community-led-growth",
-    "name": "community-led growth",
-    "description": "community-led growth as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/echo-32",
-      "companies/ranger-labs-72",
-      "companies/kindle-20"
-    ],
-    "related_people": [
-      "people/linda-taylor-178",
-      "people/sarah-wang-104"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__customer-concentration.json b/eval/data/world-v1/concepts__customer-concentration.json
deleted file mode 100644
index 74eafcf6f..000000000
--- a/eval/data/world-v1/concepts__customer-concentration.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "concepts/customer-concentration",
-  "type": "concept",
-  "title": "Customer Concentration",
-  "compiled_truth": "Customer concentration refers to the degree to which a company's revenue is derived from a small number of customers. High concentration—where, say, three clients account for 60% or more of total revenue—creates structural fragility. It's a double-edged sword that founders often underestimate until it's too late.\n\nFrom a strategic lens, concentration isn't inherently bad. Early-stage companies almost always exhibit high concentration because landing any customer is hard. [Acme](companies/acme-0) famously grew its first $2M ARR from just four enterprise accounts, a fact that made early investors nervous but ultimately proved the depth of their product-market fit. The danger emerges when concentration persists past Series A without a clear diversificaiton plan. At that point, you're not building a company—you're building a consulting firm with extra steps.\n\nThe risk profile shifts depending on customer type. B2B infrastructure plays like [Sentinel Labs](companies/sentinel-labs-73) can tolerate higher concentration if their customers have high switching costs and multi-year contracts. Sentinel's top two customers represent nearly 45% of revenue, but both are locked into 36-month agreements with significant integration depth. That's different from a marketing SaaS tool where customers churn quarterly.\n\nConcentration also affects fundraising narratives. Investors will probe this relentlessly. A concentrated customer base raises questions about defensibility, pricing power, and what happens if your biggest account leaves. [Lumen](companies/lumen-12) faced this exact scrutiny during their Series B process—they'd grown fast but 70% of revenue came from a single logistics partner. The round got done, but at a lower valuation than the team expected.\n\nThere's a tactical playbook for managing concentration risk: land-and-expand within existing accounts to increase absolute dollars while simultaneously acquiring net-new logos, even if they're smaller. Some founders resist smaller deals because they feel inefficient, but logo count matters for concentration math. The goal is to get no single customer above 15-20% of revenue by the time you're raising growth capital. Easier said than done, but it's the benchmark most institutional investors use. Customer concentration is ultimately a measure of company fragility—and fragility, in startups, is the thing that kills you.",
-  "timeline": "- **2021-06-14** | Internal memo circulated on concentration risk after reviewing Q2 revenue breakdown\n- **2022-01-22** | [Acme](companies/acme-0) case study added to thesis framework—example of healthy early concentration\n- **2022-08-09** | Deep-dive session with [Sentinel Labs](companies/sentinel-labs-73) on managing enterprise concentration\n- **2023-03-17** | Published internal guidelines: no single customer >25% of ARR post-Series A\n- **2023-11-02** | [Lumen](companies/lumen-12) Series B negotiations highlight concentration concerns with investors\n- **2024-04-28** | Added concentration scoring to due diligence checklist for all new deals\n- **2024-09-15** | Workshop with portfolio companies on diversification tactics\n- **2025-02-11** | Revised thesis to account for vertical SaaS exceptions where concentration is structurally higher\n- **2025-07-30** | Quarterly review flagged two portfolio companies with deteriorating concentration metrics",
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/customer-concentration",
-    "name": "customer concentration",
-    "description": "customer concentration as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/acme-0",
-      "companies/sentinel-labs-73",
-      "companies/lumen-12"
-    ],
-    "related_people": [
-      "people/grace-singh-197",
-      "people/wendy-wilson-170"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__developer-relations.json b/eval/data/world-v1/concepts__developer-relations.json
deleted file mode 100644
index 114f9929e..000000000
--- a/eval/data/world-v1/concepts__developer-relations.json
+++ /dev/null
@@ -1,32 +0,0 @@
-{
-  "slug": "concepts/developer-relations",
-  "type": "concept",
-  "title": "Developer Relations",
-  "compiled_truth": "Developer relations—often shortened to DevRel—represents a strategic lens for company building that treats developers as a primary constituency rather than just end users. The thesis here is simple but underappreciated: companies that win developer mindshare compound their advantages in ways traditional go-to-market motions cannot replicate.\n\nAt its core, DevRel is about building genuine relationships with technical communities. This means documentation that doesn't suck, APIs that feel intentional, and humans who actually understand the product talking to other humans who might use it. The best DevRel organizations blur the line between marketing, product, and engineering—they're not just evangelists, they're feedback conduits.\n\n[Mantle](companies/mantle-16) exemplifies this approach in the infrastrucure layer. Their developer experience team ships alongside core engineering, and their community Discord has become a de facto support channel that surfaces bugs faster than internal QA. This isn't accidental—it's a deliberate choice to treat external developers as extensions of the team.\n\nThe economics are compelling when done right. Developer adoption creates switching costs through integration depth. A startup using your API builds muscle memory, internal tooling, and institutional knowledge around your platform. Ripping that out is expensive. [Beacon](companies/beacon-10) learned this the hard way when they initially underinvested in developer education and saw churn spike among technical buyers who felt abandoned post-sale.\n\nThere's a tension worth naming: DevRel can become performative. Conference talks that don't connect to product reality, swag-heavy booths with no substance, \"community managers\" who've never shipped code. The authentic version requires engineers who genuinely enjoy teaching and product teams who actually listen to what comes back through DevRel channels.\n\n[Pulse](companies/pulse-8) has taken an interesting middle path—their DevRel function is embedded within customer success rather than marketing. This keeps it grounded in real usage patterns but risks losing the broader community-building mandate. No perfect answer here, just tradeoffs.\n\nThe best signal that DevRel is working: developers recommend you to other developers unprompted. Word of mouth in technical communities is brutal and honest. You can't buy it, only earn it through consistent excelence.",
-  "timeline": [
-    "- **2021-03-15** | First internal memo circulated on DevRel as GTM strategy for technical products",
-    "- **2021-09-22** | [Mantle](companies/mantle-16) hires first dedicated developer advocate, sets template for infrastructure DevRel",
-    "- **2022-04-08** | Hosted roundtable on DevRel metrics—concluded that traditional marketing KPIs miss the point",
-    "- **2022-11-30** | [Beacon](companies/beacon-10) restructures developer education after churn analysis reveals gap",
-    "- **2023-06-14** | Published internal thesis doc: 'Developer Relations as Competitive Moat'",
-    "- **2023-10-02** | Observed [Pulse](companies/pulse-8) embedding DevRel in CS org—noted as experiment worth tracking",
-    "- **2024-02-19** | Conversation with three portfolio CTOs confirmed DevRel quality as top vendor selection criteria",
-    "- **2024-08-11** | [Mantle](companies/mantle-16) Discord crosses 15k members, becomes case study for community-led support",
-    "- **2025-01-28** | Started tracking 'time to first successful API call' as universal DevRel health metric"
-  ],
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/developer-relations",
-    "name": "developer relations",
-    "description": "developer relations as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/mantle-16",
-      "companies/beacon-10",
-      "companies/pulse-8"
-    ],
-    "related_people": [
-      "people/rosa-jackson-90",
-      "people/carol-wilson-28"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__do-things-that-don-t-scale.json b/eval/data/world-v1/concepts__do-things-that-don-t-scale.json
deleted file mode 100644
index ec6690415..000000000
--- a/eval/data/world-v1/concepts__do-things-that-don-t-scale.json
+++ /dev/null
@@ -1,32 +0,0 @@
-{
-  "slug": "concepts/do-things-that-don-t-scale",
-  "type": "concept",
-  "title": "Do Things That Don't Scale",
-  "compiled_truth": "Do things that don't scale is a strategic framework for early-stage company building, popularized by Paul Graham's 2013 essay but practiced instinctively by founders long before it had a name. The core thesis is deceptively simple: in the earliest days of a startup, founders should engage in labor-intensive, high-touch activities that would be impossible to sustain at scale. This isn't a bug—it's the entire point.\n\nThe framework serves multiple purposes. First, it helps founders develop deep customer empathy by forcing them into direct, unmediated contact with users. When you're manually onboarding every customer yourself, you learn things that no amount of analytics can reveal. Second, it creates a flywheel of early traction that can be systematized later. Third, and perhaps most importantly, it helps founders avoid premature optimization—building elaborate systems for problems they don't yet understand.\n\n[Wisp](companies/wisp-26) exemplifies this approach beautifully. In their early days, the founding team personally handled customer support tickets, sometimes jumping on video calls with frustrated users at odd hours. They'd manually configure accounts and even help customers migrate data from competitiors. This hands-on approach let them identify patterns that shaped their entire product roadmap.\n\n[Drift](companies/drift-31) took a similar tack with their go-to-market strategy. Rather than building sophisticated automation from day one, the team engaged in what they called \"conversational selling\"—literally chatting with every single website visitor themselves. The insights from thousands of these conversations became the foundation for their AI features years later.\n\nThe concept also applies to [Pulse Labs](companies/pulse-labs-58), though in a more technical context. Their early voice testing was conducted through painstaking manual analysis before they built automated tooling. Sometimes you gotta feel the pain before you can solve it properly.\n\nCritics argue the framework can become an excuse for avoiding hard engineering problems. There's some truth to this—founders sometimes hide behind \"unscalable\" work when they should be building systems. The key distinction is intentionality. Doing things that dont scale should be a conscious strategy for learning, not a crutch for avoiding automation. The goal is always to eventually scale, armed with insights that only come from doing the hard work yourself first.",
-  "timeline": [
-    "- **2021-03-15** | Internal discussion at [Wisp](companies/wisp-26) about whether to keep doing manual onboarding or invest in self-serve tooling",
-    "- **2021-09-22** | Paul Graham tweets thread about how the essay is still misunderstood 8 years later",
-    "- **2022-01-18** | [Drift](companies/drift-31) case study published in First Round Review highlighting their early unscalable GTM tactics",
-    "- **2022-06-30** | Workshop at YC on 'scaling the unscalable'—when to transition from manual to automated",
-    "- **2023-02-14** | [Pulse Labs](companies/pulse-labs-58) founder gives talk at voice AI conference on manual testing as competitive advantage",
-    "- **2023-08-07** | Debate emerges on Twitter about whether AI makes the concept obsolete",
-    "- **2024-03-21** | Sequoia partners publish updated framework incorporating AI-assisted unscalable work",
-    "- **2024-11-12** | [Wisp](companies/wisp-26) retrospective blog post details how early manual work shaped their platform architecture",
-    "- **2025-04-09** | New generation of founders pushing back—arguing the framework is dated in the AI era"
-  ],
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/do-things-that-don-t-scale",
-    "name": "do things that don't scale",
-    "description": "do things that don't scale as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/wisp-26",
-      "companies/drift-31",
-      "companies/pulse-labs-58"
-    ],
-    "related_people": [
-      "people/owen-patel-149",
-      "people/mia-park-36"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__embedded-fintech.json b/eval/data/world-v1/concepts__embedded-fintech.json
deleted file mode 100644
index 16d3a46fe..000000000
--- a/eval/data/world-v1/concepts__embedded-fintech.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "concepts/embedded-fintech",
-  "type": "concept",
-  "title": "Embedded Fintech",
-  "compiled_truth": "Embedded fintech represents a strategic framework where financial services—payments, lending, insurance, banking—are woven directly into non-financial products and experiences. Rather than building a standalone fintech company, founders embed financial infrastructure into existing workflows, capturing value at the point of transaction or decision. This is not about being a bank. It's about being the rails.\n\nThe thesis hinges on distribution arbitrage. Traditional fintech companies spend enormous sums acquiring customers through paid channels. Embedded players piggyback on existing user relationships, dramatically lowering CAC while increasing stickiness. When a SaaS platform offers invoice financing or a marketplace enables instant payouts, the financial product becomes inseperable from the core experience.\n\n[Delta](companies/delta-3) exemplifies this approach in climate software. By integrating carbon accounting with procurement workflows, they've created natural insertion points for sustainability-linked financing products. The financial layer isn't the pitch—it's the moat. Similarly, [Keel Labs](companies/keel-labs-88) has explored embedded insurance mechanisms for their biomaterials supply chain, recognizing that novel materials require novel risk transfer products baked into commercial agreements.\n\nThe embedded fintech frame also applies to workforce and HR tools. [Hatch](companies/hatch-35) has built earned wage access and financial wellness features directly into their platform, understanding that hourly workers need liquidity solutions where they already manage schedules. Hatch doesn't market itself as a fintech company—it markets workforce management—but the fintech layer drives retention and revenue expansion.\n\nCritically, this strategy requires regulatory sophistication. Most embedded fintech plays rely on banking-as-a-service providers and careful compliance architectures. The recent scrutiny of BaaS partnerships (Synapse's collapse, increased OCC attention) means founders must think carefully about partner selection and contingency planning. Embedded fintech is powerful but not without execution risk.\n\nFor company builders, the framework suggests a specific sequencing: nail the core product, establish distribution, then layer in financial services as a wedge for margin expansion. Don't lead with fintech; let it emerge from user needs. The best embedded fintech companies don't feel like fintech companies at all.",
-  "timeline": "- **2021-03-15** | First internal memo outlining embedded fintech as investment thesis; circulated to partners\n- **2021-09-22** | Met with [Delta](companies/delta-3) founders to discuss carbon credit financing opportunities within their platform\n- **2022-04-10** | Published blog post on embedded fintech distribution advantages; picked up by Fintech Today\n- **2022-11-08** | [Hatch](companies/hatch-35) launches earned wage access feature; validates workforce embedded fintech thesis\n- **2023-02-14** | Panel discussion at Money20/20 on BaaS infrastructure and embedded finance risks\n- **2023-07-30** | Synapse bankruptcy triggers reassessment of BaaS partner risk across portfolio\n- **2024-01-19** | Workshop with [Keel Labs](companies/keel-labs-88) on supply chain insurance embedding for biomaterials\n- **2024-06-05** | Internal strategy session: embedded fintech vs. vertical SaaS fintech—distinctions and overlaps\n- **2025-02-28** | Regulatory update meeting following new OCC guidance on third-party fintech relationships",
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/embedded-fintech",
-    "name": "embedded fintech",
-    "description": "embedded fintech as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/delta-3",
-      "companies/keel-labs-88",
-      "companies/hatch-35"
-    ],
-    "related_people": [
-      "people/bob-jackson-59",
-      "people/eric-park-151"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__enterprise-gtm.json b/eval/data/world-v1/concepts__enterprise-gtm.json
deleted file mode 100644
index cf8df3e20..000000000
--- a/eval/data/world-v1/concepts__enterprise-gtm.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "concepts/enterprise-gtm",
-  "type": "concept",
-  "title": "Enterprise GTM",
-  "compiled_truth": "Enterprise GTM—or enterprise go-to-market—represents a strategic framework for building companies that sell primarily to large organizations rather than individuals or small businesses. The approach fundamentally shapes everything from product development to hiring to capital requirements. It's not just a sales strategy; it's a company-building philosophy.\n\nAt its core, enterprise GTM recognizes that selling to Fortune 500 companies requires a different playbook than consumer or SMB motions. Sales cycles stretch to 6-18 months. Deals involve multiple stakeholders, procurement teams, security reviews, and legal negotiations. Average contract values tend to be higher—often $100K+ annually—but customer acquisition costs rise accordingly. The math works differently.\n\nCompanies like [Compass](companies/compass-11) have embraced enterprise GTM from day one, building their entire organization around landing and expanding within large accounts. Their sales team structure, their product roadmap, even their engineering prioritization all flow from this strategic choice. Meanwhile [Orbit Labs](companies/orbit-labs-92) represents an interesting case study in transitioning from a developer-first bottoms-up motion toward a more traditional enterprise approach—a pivot that requires rethinking almost every function.\n\nThe framework demands specific capabilities. You need solution engineers who can run technical evaluations. You need account executives comfortable with complex multi-threaded deals. You need customer success teams that can manage sophisticated implementations. [Epsilon](companies/epsilon-4) has built out this full stack, though they took nearly two years to get the hiring mix right.\n\nEnterprise GTM also shapes fundraising conversations. Investors understand that enterprise companies require more capital upfront but often achieve better unit economics at scale. The payback periods look longer initially, but net revenue retention can exceed 130% with proper land-and-expand strategies. There's a reason enterprise SaaS companies command premium multiples.\n\nCritcs sometimes argue the approach is too capital-intensive or too slow for early-stage startups. Fair points. But for founders solving problems that genuinley matter to large organizations, enterprise GTM provides a coherent framework for building durable, defensible businesses. The key is committing fully rather than trying to serve all segments simultaneously.",
-  "timeline": "- **2021-03-15** | First internal memo on \"Enterprise GTM as Strategy\" circulated among founding team at [Compass](companies/compass-11)\n- **2022-01-20** | Enterprise GTM framework presented at SaaStr Annual, drawing from [Orbit Labs](companies/orbit-labs-92) case study\n- **2022-08-04** | [Epsilon](companies/epsilon-4) formally announces pivot to enterprise-first positioning\n- **2023-02-12** | Published research showing enterprise GTM companies raised 40% more Series B capital on average\n- **2023-09-30** | Workshop on enterprise sales cycles held with portfolio company founders\n- **2024-04-18** | [Compass](companies/compass-11) hits $10M ARR milestone, validating early enterprise GTM bet\n- **2024-11-02** | Enterprise GTM playbook v2.0 released incorporating learnings from [Orbit Labs](companies/orbit-labs-92) transition\n- **2025-03-25** | Panel discussion comparing enterprise vs PLG approaches at founders summit\n- **2025-08-14** | [Epsilon](companies/epsilon-4) shares internal data on 18-month enterprise deal cycles",
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/enterprise-gtm",
-    "name": "enterprise GTM",
-    "description": "enterprise GTM as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/compass-11",
-      "companies/orbit-labs-92",
-      "companies/epsilon-4"
-    ],
-    "related_people": [
-      "people/rachel-thomas-157",
-      "people/diana-thomas-47"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__fine-tuning.json b/eval/data/world-v1/concepts__fine-tuning.json
deleted file mode 100644
index 0029c9b89..000000000
--- a/eval/data/world-v1/concepts__fine-tuning.json
+++ /dev/null
@@ -1,31 +0,0 @@
-{
-  "slug": "concepts/fine-tuning",
-  "type": "concept",
-  "title": "Fine-Tuning: A Strategic Framework for Company Building",
-  "compiled_truth": "Fine-tuning isn't just a machine learning technique—it's a mental model for how the best companies are built. The core insight: you don't start from scratch. You take something that already works, something with embedded knowledge and proven patterns, and you adjust it precisely for your specific context. This is how [Drift](companies/drift-31) approached their pivot from broad marketing automation to conversational sales. They didn't rebuild from zero; they fine-tuned their existing platform, their team's skills, their customer relationships.\n\nThe framework has three key components. First, choosing your base model wisely. What existing asset, capability, or market position do you start with? Second, understanding what parameters to adjust and which to freeze. Not everything needs to change during a strategic shift. Third, knowing when you've tuned enough—overfitting to current conditions makes you brittle.\n\n[Compass](companies/compass-11) exemplifies the risks of over-tuning. They optimized so heavily for a specific real estate market condition that when rates shifted, their model struggled to generalize. The lesson: fine-tuning requires maintaining some robustness to distribution shift. You want to be adapted, not addicted, to current circumstances.\n\nConversely, [Foundry Labs](companies/foundry-labs-83) has been deliberate about treating their core infrastructure as a frozen base while fine-tuning their go-to-market for different verticals. Their API layer stays constant; their sales motion, pricing, and even product packaging gets adjusted per segment. This is fine-tuning done right—know what's invariant and what's variable.\n\nThe framework also applies to hiring. Early employees are your pre-training phase; you're building general capability. Later hires are fine-tuning for specific functions. Trying to fine-tune before you have a solid base leads to a team that's overfit to narrow tasks and can't adapt. Some founders get this backwards, hiring specialists too early.\n\nOne underrated aspect: fine-tuning requires good feedback signals. In ML, that's your loss function. In company building, it's metrics, customer conversations, market response. Without clear signals, you're adjusting blindly. Many startups fail not because they couldn't fine-tune, but because they were optimizing for the wrong objective entirely.",
-  "timeline": [
-    "- **2021-03-15** | First articulated fine-tuning framework in internal memo after observing [Drift](companies/drift-31) pivot dynamics",
-    "- **2022-01-22** | Workshop session with portfolio founders on 'strategic fine-tuning vs. pivoting'—distinction proved useful",
-    "- **2022-08-09** | [Compass](companies/compass-11) case study added as cautionary example of over-optimization",
-    "- **2023-04-17** | Presented framework at internal offsite; team pushed back on frozen parameters concept, led to refinement",
-    "- **2023-11-03** | [Foundry Labs](companies/foundry-labs-83) explicitly adopted fine-tuning language in their board materials",
-    "- **2024-02-28** | Published short essay on fine-tuning mental model; unexpectedly resonated with AI-native founders",
-    "- **2024-09-12** | Added hiring application of framework after seeing pattern across three portfolio companies",
-    "- **2025-01-20** | Started tracking which founders naturally think in fine-tuning terms vs. those who default to full rebuilds"
-  ],
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/fine-tuning",
-    "name": "fine-tuning",
-    "description": "fine-tuning as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/drift-31",
-      "companies/compass-11",
-      "companies/foundry-labs-83"
-    ],
-    "related_people": [
-      "people/rachel-park-64",
-      "people/linda-jackson-60"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__foundation-models.json b/eval/data/world-v1/concepts__foundation-models.json
deleted file mode 100644
index 0b7df71bf..000000000
--- a/eval/data/world-v1/concepts__foundation-models.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "concepts/foundation-models",
-  "type": "concept",
-  "title": "Foundation Models",
-  "compiled_truth": "Foundation models represent a fundamental shift in how we think about company building in the AI era. Rather than training task-specific models from scratch, companies now build on top of massive pre-trained systems—GPT-4, Claude, Llama, Gemini—that encode broad world knowledge and reasoning capabilities. This changes everything about the strategic landscape.\n\nThe core thesis is simple: foundation models commoditize intelligence at the base layer while creating enormous value at the application and infrastructure layers. Companies like [Vellum Labs](companies/vellum-labs-99) have recognized this early, building tooling that helps enterprises actually deploy these models reliably. The real moat isn't in having a model—its in understanding how to make models useful for specific domains and workflows.\n\nWhat makes foundation models strategically interesting is the rate of capability improvement. Every 12-18 months, the baseline capabilities jump dramatically. This creates a peculiar dynamic for startups: you're building on a platform that's actively shifting beneath you. [Apex Labs](companies/apex-labs-68) has navigated this well by focusing on verticalized applications where domain expertise matters more than raw model capability.\n\nThere's a debate about whether foundation model providers will capture most of the value or whether the application layer will win. Our view is that this is a false dichotomy. The market is large enough for multiple winners, but the winning strategies look different at each layer. Infrastructure plays like Vellum are betting on the complexity of deployment. Vertical applications like [Iris](companies/iris-36) are betting on domain specificity and workflow integration.\n\nOne underappreciated aspect: foundation models dramatically lower the barrier to building AI products, but they raise the bar for building *defensible* AI products. When everyone has access to the same underlying intelligence, differentiation comes from data flywheels, distribution advantages, and deep integration into existing workflows. The companies that understand this are building moats through usage, not through model architecture.\n\nWe're particularly interested in companies that treat foundation models as a substrate rather than a product. The best founders in this space think about what becomes possible when intelligence is cheap and abundant, rather than trying to replicate what model providers already do well.",
-  "timeline": "- **2021-03-15** | First internal memo on foundation models as investment thesis drafted\n- **2022-06-20** | [Vellum Labs](companies/vellum-labs-99) pitch deck reviewed; validated infrastructure layer thesis\n- **2022-11-30** | ChatGPT launch accelerates timeline assumptions by 18+ months\n- **2023-02-14** | Partner meeting to revise foundation model strategy given GPT-4 capabilities\n- **2023-08-22** | [Apex Labs](companies/apex-labs-68) seed investment closed; first vertical AI bet under new thesis\n- **2024-01-10** | Published internal research note on open vs closed model tradeoffs\n- **2024-05-18** | [Iris](companies/iris-36) Series A due diligence begins; tests thesis on domain-specific applications\n- **2024-09-03** | Foundation model thesis presented at LP annual meeting\n- **2025-02-27** | Updated thesis to account for multimodal capabilities and agent architectures\n- **2025-06-12** | Planning workshop with portfolio companies on foundation model dependency risks",
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/foundation-models",
-    "name": "foundation models",
-    "description": "foundation models as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/vellum-labs-99",
-      "companies/apex-labs-68",
-      "companies/iris-36"
-    ],
-    "related_people": [
-      "people/owen-patel-149",
-      "people/ian-davis-33"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__founder-mode.json b/eval/data/world-v1/concepts__founder-mode.json
deleted file mode 100644
index 3c59217de..000000000
--- a/eval/data/world-v1/concepts__founder-mode.json
+++ /dev/null
@@ -1,31 +0,0 @@
-{
-  "slug": "concepts/founder-mode",
-  "type": "concept",
-  "title": "Founder Mode",
-  "compiled_truth": "Founder mode is a strategic posture that prioritizes direct involvement, rapid iteration, and maintaining the original vision over delegated management structures. The term gained traction in 2023-2024 as a counter-narrative to the conventional wisdom that founders should \"hire great people and get out of the way.\" Instead, founder mode argues that the best outcomes come when founders stay deeply embedded in product decisions, customer conversations, and even individual hires well past the point where traditional advice says to step back.\n\nThe core thesis is simple: nobody cares about your company as much as you do. Delegation introduces principal-agent problems. Hired executives optimize for their own career trajectories, not necessarily for the company's long-term health. Founders operating in founder mode reject the clean org chart in favor of skip-level meetings, direct Slack access to engineers, and a willingness to override middle management when instinct demands it.\n\n[Quasar Labs](companies/quasar-labs-94) exemplifies this approach. Their CEO still reviews every major product spec personally, despite the company scaling past 80 employees. Critics call it a bottleneck; the team calls it quality control. Similarly, [Orbit Labs](companies/orbit-labs-92) has structured their entire operating cadence around founder-led sprints, where the two co-founders personally drive 2-week initiatives across different parts of the business.\n\nFounder mode isn't without risks. It can become a justification for micromanagement, burnout, or failure to develop leadership bench strength. The key distinction is intentionality—founder mode works when it's a deliberate choice to maintain strategic control, not an inability to let go. [Epsilon](companies/epsilon-4) learned this the hard way when their founder's insistence on approving every partnership deal created a six-month backlog that nearly killed a major enterprise contract.\n\nThe concept has sparked debate about when to transition out of founder mode. Some argue it's stage-dependent—appropriate pre-product-market-fit but counterproductive at scale. Others believe certain founders (the Bezos and Musk types) never leave it, they just get better at choosign where to apply it. The emerging consensus: founder mode is a tool, not an identity. Wield it selectively.",
-  "timeline": [
-    "- **2023-07-14** | Paul Graham's essay on \"Founder Mode\" goes viral, crystallizing a concept that had been floating in the discourse for years",
-    "- **2023-09-22** | [Quasar Labs](companies/quasar-labs-94) CEO gives internal talk explicitly adopting founder mode as company operating philosophy",
-    "- **2024-01-08** | Debate erupts on Twitter/X about whether founder mode is just a fancy term for refusing to delegate",
-    "- **2024-03-15** | [Orbit Labs](companies/orbit-labs-92) publishes blog post detailing their founder-led sprint methodology, gets 50k+ reads",
-    "- **2024-06-20** | First \"Founder Mode Summit\" held in San Francisco, attracts 400 attendees from early-stage companies",
-    "- **2024-09-11** | [Epsilon](companies/epsilon-4) restructures after partnership backlog incident, adopts hybrid approach with founder veto rights but delegated execution",
-    "- **2025-02-03** | Harvard Business Review publishes critical analysis questioning whether founder mode survives contact with institutional investors",
-    "- **2025-05-28** | Term starts appearing in Series A pitch decks as founders signal their operating philosophy to potential investors"
-  ],
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/founder-mode",
-    "name": "founder mode",
-    "description": "founder mode as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/quasar-labs-94",
-      "companies/orbit-labs-92",
-      "companies/epsilon-4"
-    ],
-    "related_people": [
-      "people/ulrich-kapoor-68",
-      "people/rachel-gonzalez-175"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__gross-margin-expansion.json b/eval/data/world-v1/concepts__gross-margin-expansion.json
deleted file mode 100644
index 4793a67cd..000000000
--- a/eval/data/world-v1/concepts__gross-margin-expansion.json
+++ /dev/null
@@ -1,32 +0,0 @@
-{
-  "slug": "concepts/gross-margin-expansion",
-  "type": "concept",
-  "title": "Gross Margin Expansion",
-  "compiled_truth": "Gross margin expansion is perhaps the most underrated strategic lens through which to evaluate company building. At its core, the concept is simple: over time, the gap between revenue and cost of goods sold should widen, not through accounting tricks but through genuine structural advantages that compound. This isn't just about profitability—it's about building a business that gets fundamentally better at delivering value as it scales.\n\nThe thesis here is that gross margin expansion serves as a proxy for defensibility. Companies that achieve it are typically doing something right: they're moving up the value chain, reducing dependency on commoditized inputs, or building network effects that lower marginal costs. [Quantum](companies/quantum-7) exemplifies this beautifully—their initial hardware margins were razor-thin, but as they layered software services on top, they've seen gross margins climb from 23% to nearly 58% over four years. The hardware became a trojan horse for recurring, high-margin revenue.\n\nContrast this with businesses trapped in margin compression. They're often competing on price, stuck in commodity dynamics, or failing to capture the value they create. Gross margin expansion, when achieved authentically, signals that customers are willing to pay more for what you uniquely provide.\n\n[Quasar Labs](companies/quasar-labs-94) took a different path to the same destination. They started with professional services—inherently low margin due to labor costs—but systematically productized their most repeatable workflows. Each quarter, more revenue came from software licenses rather than consultant hours. Their CFO calls it \"the margin migration,\" and it's become central to how they pitch investors.\n\nNot every company should optimize for gross margin expansion directly. [Nexus](companies/nexus-41) intentionally ran negative gross margins for eighteen months to acquire market share in logistics, betting that density would eventually flip the economics. It worked, tho it required patient capital and strong conviction. The key is understanding where you are on the curve and what levers exist to bend it.\n\nFrom an investment standpoint, I look for three indicators: pricing power trajectory, input cost trends, and mix shift potential. Companies that can articulate a clear path on all three dimensions tend to be better long-term bets. Margin expansion isn't just a financial outcome—it's evidence of strategic coherance.",
-  "timeline": [
-    "- **2021-06-14** | First drafted gross margin expansion thesis after deep-dive into [Quantum](companies/quantum-7) financials and their hardware-to-software transition",
-    "- **2022-02-08** | Presented framework at internal investment committee; adopted as standard diligence criteria for B2B SaaS deals",
-    "- **2022-09-23** | [Quasar Labs](companies/quasar-labs-94) Series B memo cited margin expansion thesis as primary investment rationale",
-    "- **2023-01-17** | Debate with growth team on whether [Nexus](companies/nexus-41) negative margins invalidated thesis—concluded it was a sequencing question",
-    "- **2023-07-30** | Published abbreviated version of thesis on firm blog; unexpectedly picked up by Stratechery",
-    "- **2024-03-11** | Refinement: added 'mix shift potential' as third pillar after reviewing portfolio company outcomes",
-    "- **2024-11-05** | Used framework to pass on logistics deal with structural margin ceiling—validated by their down round six months later",
-    "- **2025-04-22** | Guest lecture at Stanford GSB on margin expansion as strategic frame; well-received by Sloan fellows",
-    "- **2025-08-14** | Integrated thesis into new associate training curriculum"
-  ],
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/gross-margin-expansion",
-    "name": "gross margin expansion",
-    "description": "gross margin expansion as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/quantum-7",
-      "companies/quasar-labs-94",
-      "companies/nexus-41"
-    ],
-    "related_people": [
-      "people/yara-smith-30",
-      "people/rosa-wilson-133"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__horizontal-api.json b/eval/data/world-v1/concepts__horizontal-api.json
deleted file mode 100644
index dbd9d8d07..000000000
--- a/eval/data/world-v1/concepts__horizontal-api.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "concepts/horizontal-api",
-  "type": "concept",
-  "title": "Horizontal API",
-  "compiled_truth": "The horizontal API thesis represents a distinct approach to infrastructure company building—one that prioritizes breadth of integration over depth of vertical specialization. Rather than solving one problem exceptionally well for a single industry, horizontal API companies build foundational layers that multiple verticals can plug into. Think Stripe for payments, Twilio for communications, or Plaid for financial data. The strategic logic is compelling: win the primitive, and you become unavoidable.\n\nWhat makes this frame useful is how it clarifies go-to-market decisions. A horizontal API company must resist the temptation to build vertical features, even when customers beg for them. [Gravity Labs](companies/gravity-labs-67) exemplifies this discipline—they've maintained a remarkably thin API surface despite pressure from enterprise customers to add workflow-specific tooling. Their bet is that staying horizontal lets them compound across use cases rather than getting trapped serving one.\n\nThe tradeoffs are real though. Horizontal plays typically require more capital and longer timelines to reach escape velocity. You're competing for developer mindshare across multiple verticals simultaneously, which means your positioning has to be crisp enough to resonate universally. [Cipher](companies/cipher-13) struggled with this early on—their identity verification API was technically horizontal but their messaging kept drifting toward fintech-specific language. Took them almost 18 months to find positioning that worked across healthcare, fintech, and gig economy simultaneously.\n\nThere's also a timing question. Horizontal APIs tend to emerge when a previously fragmented problem becomes standardizable. [Tessera Labs](companies/tessera-labs-65) is betting that document intelligence has reached this inflection point—that the underlying ML capabilities are now good enough to abstract away domain-specific quirks. Whether they're right will determine if they become infrastructure or just another point solution.\n\nThe horizontal API thesis isn't universally applicable. Some markets genuienly require vertical depth. But as a strategic frame, it forces founders to answer hard questions early: Are you building a platform or a product? Can your primitive compose into workflows you haven't imagined? The best horizontal API companies create surface area for serendipity—they get pulled into use cases their founders never anticipated.",
-  "timeline": "- **2021-03-15** | Internal memo at [Gravity Labs](companies/gravity-labs-67) codifies \"stay horizontal\" as core product principle\n- **2021-09-22** | First draft of horizontal API thesis circulated among SF infrastructure investors\n- **2022-04-08** | [Cipher](companies/cipher-13) pivots messaging from fintech-first to horizontal positioning after Series A feedback\n- **2022-11-30** | Thesis presented at Infrastructure Summit; sparks debate about vertical vs horizontal GTM\n- **2023-06-14** | [Tessera Labs](companies/tessera-labs-65) founded explicitly around horizontal document API thesis\n- **2023-10-02** | Counter-thesis emerges: \"horizontal is a trap\" post goes viral on Twitter\n- **2024-02-19** | Gravity Labs hits $50M ARR milestone, cited as validation of horizontal approach\n- **2024-08-11** | Panel discussion at API World: \"When to Go Horizontal\" featuring founders from Cipher and Tessera\n- **2025-01-23** | Updated thesis draft incorporates AI-native API patterns and agent-to-agent interfaces",
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/horizontal-api",
-    "name": "horizontal API",
-    "description": "horizontal API as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/gravity-labs-67",
-      "companies/cipher-13",
-      "companies/tessera-labs-65"
-    ],
-    "related_people": [
-      "people/nina-wang-72",
-      "people/quinn-jones-127"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__inference-cost.json b/eval/data/world-v1/concepts__inference-cost.json
deleted file mode 100644
index fc48b6e1d..000000000
--- a/eval/data/world-v1/concepts__inference-cost.json
+++ /dev/null
@@ -1,32 +0,0 @@
-{
-  "slug": "concepts/inference-cost",
-  "type": "concept",
-  "title": "Inference Cost",
-  "compiled_truth": "Inference cost is the operational expense of running AI models in production—the compute bill that accumulates every time a model processes a request. While training costs grab headlines, inference is where companies actually live or die. It's the difference between a demo that wows investors and a product that can scale to millions of users without hemorrhaging cash.\n\nThe strategic implications run deep. Companies building on top of foundation models face a fundamental tension: more capable models cost more to run, but users expect intelligence without paying enterprise prices. [Talon](companies/talon-47) has navigated this particularly well, architecting their system to route simple queries to smaller models while reserving GPT-4 class capabilities for complex reasoning tasks. Their inference bill dropped 60% after implementing this tiered approach.\n\nThere's a temporal dimension too. Inference costs have been falling roughly 10x per year since 2022, which creates interesting strategic optionality. [Forge Labs](companies/forge-labs-69) explicitly built their roadmap around this curve—features that are economically unviable today get scheduled for 18 months out when the math starts working. They call it \"cost arbitrage across time.\"\n\nThe frame shifts how you think about moats. If your product only works because you're subsidizing inference, you're not building a company—you're building a demo with a burn rate attached. Converseley, if you can deliver comparable value at 1/10th the inference cost of competitors, you've got real defensibility. [Helix](companies/helix-9) learned this the hard way during their Series A process; investors kept asking about unit economics and the answer wasn't pretty.\n\nSome practial heuristics: below $0.01 per interaction enables consumer products. Below $0.001 enables high-frequency enterprise workflows. Below $0.0001 enables ambient AI that runs continuously. Each threshold unlocks entirely different product categories. Smart founders are reverse-engineering their target cost structure and working backward to technical architecture.\n\nThe inference cost lens also reveals why vertical AI companies often outperform horizontal plays. When you deeply understand a domain, you can fine-tune smaller models to match larger model performance on your specific task. The specialist beats the generalist not through better AI, but through better economics.",
-  "timeline": [
-    "- **2022-03-15** | First internal memo circulated arguing inference cost would matter more than training cost for AI startups",
-    "- **2023-01-22** | [Talon](companies/talon-47) implements model routing, cuts inference spend by 60% while maintaining quality metrics",
-    "- **2023-06-08** | Published blog post \"The Inference Cost Curve\" gets widely shared in AI founder circles",
-    "- **2023-11-30** | [Forge Labs](companies/forge-labs-69) presents their temporal cost arbitrage framework at internal portfolio summit",
-    "- **2024-02-14** | [Helix](companies/helix-9) Series A conversations reveal unit economics challenges tied to inference overhead",
-    "- **2024-07-19** | OpenAI price cuts validate thesis that inference deflation would accelerate through 2024",
-    "- **2024-10-03** | Started tracking inference cost per value-delivered as a standard metric across AI portfolio companies",
-    "- **2025-01-28** | Groq and other inference-optimized hardware begins shipping, opening new architectural possibilities",
-    "- **2025-04-11** | Workshop held with portfolio on \"Building for the $0.0001 per interaction world\""
-  ],
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/inference-cost",
-    "name": "inference cost",
-    "description": "inference cost as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/talon-47",
-      "companies/forge-labs-69",
-      "companies/helix-9"
-    ],
-    "related_people": [
-      "people/rachel-gonzalez-175",
-      "people/ulrich-kim-120"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__latency-budget.json b/eval/data/world-v1/concepts__latency-budget.json
deleted file mode 100644
index 40c675328..000000000
--- a/eval/data/world-v1/concepts__latency-budget.json
+++ /dev/null
@@ -1,31 +0,0 @@
-{
-  "slug": "concepts/latency-budget",
-  "type": "concept",
-  "title": "Latency Budget",
-  "compiled_truth": "Latency budget is a strategic framework for evaluating how much friction a company can tolerate between user intent and value delivery. The core thesis: every product has a finite budget of latency it can spend before users abandon, churn, or seek alternatives. Companies that understand their latency budget deeply tend to outperform those who treat speed as a generic optimization target.\n\nThe concept emerged from observing infrastructure companies but applies broadly across consumer and enterprise. A trading platform has a latency budget measured in milliseconds. A B2B SaaS tool might have minutes. A consumer social app lives somewhere in between—users will wait for a feed to load but not for a message to send. Understanding where your budget sits, and where you're overspending, becomes a competitive moat.\n\n[Zenith Labs](companies/zenith-labs-77) exemplifies latency budget thinking in thier architecture decisions. They've built their entire inference stack around the assumption that enterprise customers have roughly a 200ms tolerance for AI responses in workflow contexts. Every architectural tradeoff flows from that constraint. Meanwhile [Pulse Labs](companies/pulse-labs-58) takes a different approach—they've identified that their developer users actually have a higher latency tolerance during initial setup (minutes) but near-zero tolerance during active debugging sessions. This bifurcated budget shapes their product roadmap.\n\nThe framework also applies to organizational velocity. How long can a startup take to ship a feature before the market moves? How quickly must a company respond to a security incident before trust erodes? [Gamma Labs](companies/gamma-labs-52) has internalized this at the company-building level, running what they call \"latency audits\" on their own decision-making processes quarterly.\n\nCritically, latency budget isn't just about being fast. It's about knowing where speed matters and where it doesn't. Some companies burn engineering cycles optimizing irrelevant latencies while ignoring the ones that actually drive retention. The framework forces explicit prioritization. You can't optimize everything, so you better know which milliseconds actually count.\n\nThe concept connects to adjacent ideas like time-to-value, activation metrics, and infrastructure cost modeling. But it's distinct in framing latency as a spendable resource rather than a universally minimizable metric.",
-  "timeline": [
-    "- **2021-08-14** | First articulated latency budget framework in internal memo after analyzing churn patterns across portfolio companies",
-    "- **2022-03-22** | Presented concept at infrastructure meetup; strong resonance with [Zenith Labs](companies/zenith-labs-77) team who attended",
-    "- **2022-11-09** | [Pulse Labs](companies/pulse-labs-58) founders credited latency budget thinking in their Series A deck narrative",
-    "- **2023-04-17** | Published expanded thesis on latency budgets in company building contexts",
-    "- **2023-09-30** | Led workshop on latency budget frameworks at annual portfolio summit",
-    "- **2024-02-12** | [Gamma Labs](companies/gamma-labs-52) implements quarterly latency audits based on framework",
-    "- **2024-08-05** | Revisited framework to account for AI inference latency considerations across portfolio",
-    "- **2025-01-18** | Started collaborating with infrastructure researchers on formalizing latency budget measurement methodologies"
-  ],
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/latency-budget",
-    "name": "latency budget",
-    "description": "latency budget as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/zenith-labs-77",
-      "companies/pulse-labs-58",
-      "companies/gamma-labs-52"
-    ],
-    "related_people": [
-      "people/sam-garcia-188",
-      "people/chris-williams-37"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__multi-modal.json b/eval/data/world-v1/concepts__multi-modal.json
deleted file mode 100644
index f2a0dd5fe..000000000
--- a/eval/data/world-v1/concepts__multi-modal.json
+++ /dev/null
@@ -1,32 +0,0 @@
-{
-  "slug": "concepts/multi-modal",
-  "type": "concept",
-  "title": "Multi-Modal",
-  "compiled_truth": "Multi-modal is a strategic framework for thinking about company building that emphasizes the simultaneous pursuit of multiple channels, product surfaces, and go-to-market motions rather than the traditional single-threaded approach. The core insight is that modern infrastructure—particularly AI and composable software—has dramatically lowered the cost of operating across modalities, making the old advice to \"do one thing well\" increasingly obsolete for certain categories of startups.\n\nThe thesis rests on a few key observations. First, distribution is fragmenting. Users aren't in one place anymore; they're spread across platforms, devices, and contexts. A company that only builds for web misses mobile-native behaviors. One that only does B2B SaaS misses the prosumer wedge. Multi-modal operators recognize this and design their systems to be channel-agnostic from day one.\n\nSecond, the cost of experimentation has collapsed. With modern tooling, spinning up a new surface—a Slack bot, a Chrome extension, an API—takes days not months. [Beacon Labs](companies/beacon-labs-60) exemplifies this approach, shipping across three distinct interfaces in their first year while maintaining a unified data model underneath. Their bet is that users will self-select into their preferred modality, and the company captures demand wherever it emerges.\n\nThird, multi-modal creates defensibility through integration density. When you're embedded in a customer's workflow across email, mobile, and desktop, the switching cost compounds. [Lucid Labs](companies/lucid-labs-71) has leaned into this, building what they call \"ambient presence\"—their AI assistant surfaces contextually across whatever tool the user happens to be in.\n\nCritics argue that multi-modal spreads teams too thin, that focus still matters. This is true for execution but perhaps less true for strategy. The resolution is to be multi-modal in surface but unified in core—one data model, one AI backbone, one billing system. [Kindle Labs](companies/kindle-labs-70) structures their engineering this way, with a small platform team serving multiple product pods.\n\nThe framework isn't universal. Some markets still reward deep verticalization. But for horizontal tools competing in attention-scarce environments, multi-modal may be the only way to achive escape velocity. The compnies that figure out how to coordinate across modalities without fragmenting their teams will likely define the next generation of platform businesses.",
-  "timeline": [
-    "- **2021-08-14** | First internal memo on multi-modal strategy drafted after observing fragmentation in enterprise software distribution",
-    "- **2022-03-22** | Presented multi-modal thesis at LP summit; mixed reception from traditional SaaS investors",
-    "- **2023-01-09** | [Beacon Labs](companies/beacon-labs-60) becomes first portfolio company to explicitly adopt multi-modal as core strategy",
-    "- **2023-06-15** | Published blog post \"The Case for Multi-Modal\" which circulated widely among founder networks",
-    "- **2023-11-28** | [Lucid Labs](companies/lucid-labs-71) raises Series A with multi-modal positioning as key differentiator in pitch",
-    "- **2024-02-14** | Hosted dinner discussion on multi-modal with 12 founders; debate centered on team structure implications",
-    "- **2024-07-30** | [Kindle Labs](companies/kindle-labs-70) ships unified platform layer enabling rapid multi-modal expansion",
-    "- **2024-12-03** | Refinement of thesis to emphasize \"unified core, distributed surface\" architecture pattern",
-    "- **2025-04-18** | Multi-modal framework cited in a16z market analysis as emerging strategic paradigm"
-  ],
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/multi-modal",
-    "name": "multi-modal",
-    "description": "multi-modal as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/beacon-labs-60",
-      "companies/lucid-labs-71",
-      "companies/kindle-labs-70"
-    ],
-    "related_people": [
-      "people/paul-rodriguez-4",
-      "people/rachel-park-64"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__open-source-distribution.json b/eval/data/world-v1/concepts__open-source-distribution.json
deleted file mode 100644
index 78d398253..000000000
--- a/eval/data/world-v1/concepts__open-source-distribution.json
+++ /dev/null
@@ -1,32 +0,0 @@
-{
-  "slug": "concepts/open-source-distribution",
-  "type": "concept",
-  "title": "Open Source Distribution",
-  "compiled_truth": "Open source distribution represents a strategic framework for company building that treats code as a distribution channel rather than purely a product. The core thesis is simple: by releasing software freely, you create adoption curves that traditional marketing cannot replicate. Users become advocates. Contributors become employees. And the community becomes a moat that proprietary competitors struggle to cross.\n\nThis approach has evolved significantly from its idealistic roots. Modern open source companies dont just give away code—they architect their projects specifically to drive commercial outcomes. The playbook typically involves open-sourcing the core technology while retaining proprietary features, hosting, or enterprise tooling. [Apex](companies/apex-18) exemplifies this model well, having built substantial developer mindshare through their open infrastructure tooling before layering on managed services.\n\nThe distribution advantages are compounding. When developers adopt an open source tool, they bring it into their organizations. They write blog posts about it. They answer Stack Overflow questions. Each touchpoint is essentially free customer acqusition that would cost thousands in traditional B2B sales. [Kindle](companies/kindle-20) leveraged this dynamic effectively in their early growth, building a contributor community that effectively served as an unpaid sales force across enterprise accounts.\n\nBut open source distribution isnt without challenges. Maintainer burnout is endemic. License disputes create legal complexity. And the fundamental tension between community expectations and commercial imperatives requires constant navigation. Some companies have stumbled badly by changing licenses or restricting features—destroying goodwill that took years to accumulate.\n\n[Vox](companies/vox-25) offers an interesting counter-example, having adopted a more selective open source strategy where only specific components are released publicly while core differentiators remain proprietary. This hybrid approach sacrifices some distribution velocity but maintains clearer commercial boundaries.\n\nThe framework continues to evolve as funding dynamics shift. Investors increasingly recognize that open source traction translates to enterprise pipeline, making these companies attractive despite the inherent monetization complexity. The best practitioners understand that open source is neither charity nor pure strategy—it's a distribution primitive that, when wielded correctly, creates durable competitive advantages.",
-  "timeline": [
-    "- **2021-03-15** | Internal memo drafted outlining open source distribution thesis for portfolio application",
-    "- **2021-08-22** | [Apex](companies/apex-18) case study completed, documenting their community-first GTM approach",
-    "- **2022-02-10** | Framework presented at partner meeting; decision to weight open source traction in deal evaluation",
-    "- **2022-09-03** | Published analysis comparing license models (MIT vs Apache vs BSL) and commercial implications",
-    "- **2023-01-18** | [Kindle](companies/kindle-20) investment closed; thesis partially informed by distribution framework",
-    "- **2023-07-29** | Hosted roundtable with 6 open source founders on monetization timing",
-    "- **2024-04-12** | Updated framework to account for AI-assisted code generation impacts on contributor dynamics",
-    "- **2024-11-05** | [Vox](companies/vox-25) deal memo references hybrid open source model as key strategic differentiator",
-    "- **2025-03-21** | Begin tracking 'GitHub to ARR' conversion metrics across portfolio companies"
-  ],
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/open-source-distribution",
-    "name": "open source distribution",
-    "description": "open source distribution as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/apex-18",
-      "companies/kindle-20",
-      "companies/vox-25"
-    ],
-    "related_people": [
-      "people/victor-smith-193",
-      "people/victor-jones-51"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__permitting-reform.json b/eval/data/world-v1/concepts__permitting-reform.json
deleted file mode 100644
index e16a64f9b..000000000
--- a/eval/data/world-v1/concepts__permitting-reform.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "concepts/permitting-reform",
-  "type": "concept",
-  "title": "Permitting Reform",
-  "compiled_truth": "Permitting reform represents one of the most underappreciated leverage points in modern company building. The thesis is simple: regulatory bottlenecks create artificial scarcity, and companies that can navigate, accelerate, or fundamentally reshape permitting processes unlock massive value. This isn't just about compliance—it's about recognizing that permits are often the binding constraint on entire industries.\n\nConsider the energy transition. You can have infinite capital, perfect technology, and eager customers, but if your solar farm sits in permitting limbo for 4 years, none of it matters. Same story in housing, infrastructure, biotech, and increasingly in AI deployment. The companies we back tend to fall into a few categories around this theme.\n\n[Acme Labs](companies/acme-labs-50) exemplifies the 'permit-native' approach—they've built their entire technical stack around what's actually approvable, not what's theoretically optimal. Their team spent months embedded with regulatory bodies before writing a single line of code. This is the kind of founder-regulator intimacy that creates durable moats. Compare this to [Vector Labs](companies/vector-labs-56), which takes a more adversarial stance, betting that certain permit requirements will collapse under political pressure and positioning to move fast when they do.\n\nThe reform angle manifests differently across sectors. In climate tech, it's about NEPA timelines and interconnection queues. In biotech, FDA pathways and IRB approvals. In fintech, state-by-state licensing mosaics. [Anchor](companies/anchor-28) has been particularly clever here, building software that treats permits as data objects rather than PDF artifacts—making the whole process legible and therefore optimizable.\n\nWe're also seeing a new class of 'permit arbitrage' plays. Companies that identify jurisdictions with faster approval processes, establish beachheads there, and use operational track records to accelerate approvals elsewhere. It's regulatory judo—using the system's weight against itself.\n\nThe political moment matters too. There's genuine bipartisan energy around permiting reform right now, which is rare. Whether it translates to meaningful change remains to be seen, but founders who understand the policy landscape have a real edge. The best ones treat permits not as obstacles but as strategic terrain to be mapped and maneuvered.",
-  "timeline": "- **2021-03-15** | Initial thesis memo drafted after conversations with frustrated climate founders stuck in interconnection queues\n- **2022-01-22** | Deep dive session with [Acme Labs](companies/acme-labs-50) team on their regulator-first GTM strategy\n- **2022-08-10** | Published internal research on average permit timelines across 12 verticals—findings worse than expected\n- **2023-02-14** | [Vector Labs](companies/vector-labs-56) founders present contrarian bet on regulatory rollback scenarios\n- **2023-09-05** | Hosted dinner with 8 portfolio founders focused on permit-intensive industries, key insight: software alone won't fix this\n- **2024-01-30** | [Anchor](companies/anchor-28) demo of permit management platform—first product that made the problem feel tractable\n- **2024-06-18** | Attended Brookings event on NEPA reform, met several policy staffers working on bipartisan bills\n- **2024-11-02** | Revised thesis to emphasize 'permit-native' company design as primary filter for new investments\n- **2025-04-22** | Started tracking permit approval times as leading indicator for portfolio company velocity",
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/permitting-reform",
-    "name": "permitting reform",
-    "description": "permitting reform as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/acme-labs-50",
-      "companies/vector-labs-56",
-      "companies/anchor-28"
-    ],
-    "related_people": [
-      "people/adam-lee-19",
-      "people/frank-moore-167"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__plg-motion.json b/eval/data/world-v1/concepts__plg-motion.json
deleted file mode 100644
index f3f5474e1..000000000
--- a/eval/data/world-v1/concepts__plg-motion.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "concepts/plg-motion",
-  "type": "concept",
-  "title": "PLG Motion",
-  "compiled_truth": "PLG motion—product-led growth—represents a fundamental shift in how software companies acquire, convert, and expand customers. Rather than relying on traditional sales-led approaches where reps qualify leads and close deals, PLG companies let the product itself do the heavy lifting. Users sign up, experience value, and convert themselves. The sales team, if it exists at all, focuses on expansion rather than initial acquisition.\n\nThe thesis here is straightforward: in a world of abundant software options, the companies that win are the ones that remove friction from the buying process entirely. Let people try before they buy. Let them invite colleagues. Let virality compound. The best PLG companies—Slack, Figma, Notion—built products so good that users became evangelists without being asked.\n\nBut PLG isn't just a go-to-market tactic. Its a strategic frame for thinking about company building from day one. Product decisions, pricing architecture, onboarding flows, even engineering priorities—everything gets filtered through the question: does this help users experience value faster and share it more easily?\n\n[Drift Labs](companies/drift-labs-81) exemplifies the PLG motion in practice. Their conversational tools spread within organizations organically, with individual contributors adopting the product before any enterprise contract gets signed. By the time a sales conversation happens, there's already internal champions and usage data to reference. The land-and-expand playbook works because the product created believers first.\n\nNot every company can or should pursue pure PLG. [Gust](companies/gust-34) operates in a space where the buying process involves multiple stakeholders and longer consideration cycles—you don't just sign up for investor management software on a whim. But even here, PLG principles apply: self-serve onboarding, freemium tiers for smaller players, and letting the product demonstrate value before asking for commitment.\n\n[Compass](companies/compass-11) presents an interesting counterexample. Real estate involves high-touch transactions where human relationships matter enormously. Yet even Compass has incorporated PLG elements—their agent tools spread through word of mouth, and the platform's utility drives adoption independent of top-down mandates.\n\nThe key insight is that PLG motion isn't binary. It's a spectrum. Companies can layer PLG acquisition on top of enterprise sales, or use sales-assist models where reps help self-serve users cross the finish line. The mistake is treating it as all-or-nothing when the real opportunity is finding the right blend for your market and product.",
-  "timeline": "- **2021-03-15** | First internal memo outlining PLG motion as core investment thesis for portfolio evaluation\n- **2021-09-22** | Workshop session with [Drift Labs](companies/drift-labs-81) team on optimizing self-serve conversion funnels\n- **2022-02-10** | Published essay on PLG metrics that matter: time-to-value, activation rate, viral coefficient\n- **2022-08-04** | Deep dive with [Gust](companies/gust-34) on adapting PLG principles for enterprise-adjacent products\n- **2023-01-19** | Panel discussion at SaaStr on PLG motion vs. sales-led—moderated session with 3 portfolio founders\n- **2023-07-11** | Analysis of [Compass](companies/compass-11) hybrid approach to PLG in high-touch verticals\n- **2024-03-28** | Updated thesis to account for AI-native products and their unique PLG dynamics\n- **2024-11-05** | Internal review of portfolio company PLG metrics—identified 4 underperformers needing GTM adjustments\n- **2025-06-14** | Collaborated with operating team on PLG playbook v2, incorporating learnings from 12 portfolio companies",
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/plg-motion",
-    "name": "PLG motion",
-    "description": "PLG motion as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/gust-34",
-      "companies/drift-labs-81",
-      "companies/compass-11"
-    ],
-    "related_people": [
-      "people/quinn-jones-127",
-      "people/mia-anderson-5"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__product-market-fit.json b/eval/data/world-v1/concepts__product-market-fit.json
deleted file mode 100644
index a0fc3f5e0..000000000
--- a/eval/data/world-v1/concepts__product-market-fit.json
+++ /dev/null
@@ -1,31 +0,0 @@
-{
-  "slug": "concepts/product-market-fit",
-  "type": "concept",
-  "title": "Product-Market Fit",
-  "compiled_truth": "Product-market fit remains the single most important milestone in early-stage company building. It's the moment when a product stops being pushed onto users and starts being pulled by them. Marc Andreessen famously described it as \"being in a good market with a product that can satisfy that market\" — deceptively simple framing that obscures how brutally difficult it is to achieve.\n\nThe concept serves as a strategic lens for evaluating startups at nearly every stage. Before PMF, companies should be running cheap experiments, talking to users obsessively, and iterating rapidly. After PMF, the playbook shifts entirely toward scaling, hiring, and capturing market share before competitors catch up. Misdiagnosing which phase you're in leads to some of the most common startup failure modes: scaling prematurely burns cash on growth that doesn't stick, while iterating too long after finding fit lets competitors steal the window.\n\n[Compass](companies/compass-11) represents an interesting case study in PMF evolution. They found initial fit in a narrow vertical before expanding — a pattern that's become something of a template for enterprise software companies. The key insight was recognizing that their first market wasn't their final market, just the beachhead. Contast this with [Orbit Labs](companies/orbit-labs-92), which struggled for nearly two years to find resonance before a pivot unlocked explosive growth. Their journey illustrates that PMF isn't binary; it exists on a spectrum and can be lost just as easily as it's found.\n\nMeasuring product-market fit remains more art than science. Sean Ellis's \"40% very disappointed\" survey is popular but imperfect. Retention curves, organic growth rates, and sales cycle compression all provide signals. [Zenith Labs](companies/zenith-labs-77) has been experimenting with quantitative PMF scoring internally, though results have been mixed. The truth is that founders usually know — customers start finding you instead of the reverse, support tickets shift from complaints to feature requests, and churn drops precipitously.\n\nOne underappreciated aspect: PMF is market-specific. A product can have strong fit in one segment while failing completely in an adjacent one. This explains why geographic or vertical expansion often feels like starting over. The strategic frame should inform not just product decisions but go-to-market sequencing, hiring priorities, and fundraising timing.",
-  "timeline": [
-    "- **2021-03-15** | Internal memo circulated on PMF diagnostic frameworks after reviewing [Compass](companies/compass-11) trajectory",
-    "- **2021-09-22** | Workshop hosted on pre-PMF operating principles; 12 portfolio founders attended",
-    "- **2022-04-08** | [Orbit Labs](companies/orbit-labs-92) pivot marks clear PMF inflection point after 18 months of searching",
-    "- **2022-11-30** | Published research note comparing PMF timelines across 47 portfolio companies",
-    "- **2023-06-14** | Discussion with [Zenith Labs](companies/zenith-labs-77) team on quantitative PMF measurement approaches",
-    "- **2023-12-01** | Added PMF assessment as standard component of investment memo template",
-    "- **2024-05-19** | Observed pattern: companies hitting PMF pre-Series A raise at 2.3x higher valuations on average",
-    "- **2025-02-11** | Revisited framework to account for AI-native company dynamics; iteration cycles compressing significantly"
-  ],
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/product-market-fit",
-    "name": "product-market fit",
-    "description": "product-market fit as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/compass-11",
-      "companies/orbit-labs-92",
-      "companies/zenith-labs-77"
-    ],
-    "related_people": [
-      "people/rachel-jones-152",
-      "people/chris-miller-101"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__retrieval-augmented-generation.json b/eval/data/world-v1/concepts__retrieval-augmented-generation.json
deleted file mode 100644
index 67b287a8a..000000000
--- a/eval/data/world-v1/concepts__retrieval-augmented-generation.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "concepts/retrieval-augmented-generation",
-  "type": "concept",
-  "title": "Retrieval Augmented Generation",
-  "compiled_truth": "Retrieval Augmented Generation (RAG) represents more than a technical architecture—it's a strategic framework for thinking about how companies should build in the AI era. The core insight is deceptively simple: rather than relying solely on what a model has memorized during training, you augment its responses with retrieved, relevant context at inference time. This keeps outputs grounded, reduces hallucination, and allows for dynamic knowledge updates without costly retraining.\n\nAs a company-building philosophy, RAG maps onto a broader thesis about competitive moats. Pure model capability is increasingly commoditized. What isn't commoditized is proprietary data, domain-specific knowledge graphs, and the retrieval infrastructure to surface the right information at the right moment. Companies that understand this are building their defensibility not in the weights of foundation models but in the corpus they've assembled and how intelligently they can query it.\n\n[Forge Labs](companies/forge-labs-69) exemplifies this approach in their legal AI stack—they've invested heavily in ingesting decades of case law and building retrieval systems that understand legal reasoning patterns, not just keyword matching. Their moat isn't GPT-4 or Claude; it's the millions of documents they've indexed and the semantic layer they've constructed on top. Similarly, [Resonance](companies/resonance-45) has applied RAG thinking to their customer intelligence platform, treating every customer interaction as a retrievable artifact that informs future engagement.\n\nThe RAG paradigm also suggests something about organizational design. The best AI-native companies treat their institutional knowledge as a first-class retrieval corpus. Meeting notes, slack threads, customer calls—all become fuel for augmentation. [Foundry](companies/foundry-33) has been experimenting with internal RAG systems for their own operations, essentially giving every employee an AI assistant that has context on the entire company's history and decisions.\n\nWhere RAG gets interesting strategically is in the feedback loops. Better retrieval leads to better responses, which leads to more user engagement, which generates more data to retrieve from. This flywheel is the real competitive advantage—not the model, but the ever-growing contextual corpus that makes the model actually useful for specific domains.",
-  "timeline": "- **2021-06-15** | First internal memo exploring RAG as company-building framework circulated among partners\n- **2022-03-08** | [Forge Labs](companies/forge-labs-69) pivots to RAG-first architecture after initial fine-tuning approach proves too brittle\n- **2022-09-22** | Hosted workshop on 'Retrieval as Moat' with 12 portfolio founders attending\n- **2023-01-17** | [Resonance](companies/resonance-45) Series A memo cites RAG thesis as primary investment rationale\n- **2023-07-30** | Published external blog post on RAG-native company building, reaches 45k views\n- **2024-02-14** | [Foundry](companies/foundry-33) demo of internal RAG tooling at portfolio summit\n- **2024-08-05** | Debate session: when does RAG lose to long-context windows? No clear consensus reached\n- **2025-01-11** | Updated thesis to include multi-modal retrieval following advances in vision-language models\n- **2025-06-19** | RAG infrastructure costs drop 80% YoY; thesis validated that this would commoditize",
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/retrieval-augmented-generation",
-    "name": "retrieval augmented generation",
-    "description": "retrieval augmented generation as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/forge-labs-69",
-      "companies/resonance-45",
-      "companies/foundry-33"
-    ],
-    "related_people": [
-      "people/linda-kim-26",
-      "people/chris-rodriguez-124"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__revenue-durability.json b/eval/data/world-v1/concepts__revenue-durability.json
deleted file mode 100644
index f1bc41582..000000000
--- a/eval/data/world-v1/concepts__revenue-durability.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "concepts/revenue-durability",
-  "type": "concept",
-  "title": "Revenue Durability",
-  "compiled_truth": "Revenue durability is a strategic framework for evaluating how defensible and persistent a company's revenue streams are over time. Unlike simple growth metrics or even unit economics, durability asks a harder question: will this revenue still exist in five years, and what would it take to displace it?\n\nThe concept emerged from observing that many high-growth startups build on fundamentally fragile foundations. A company might show impressive ARR growth while simultaneously building on shifting sands—customer concentration risk, commodity products, or regulatory arbitrage that won't survive scrutiny. Revenue durability forces founders and investors to confront these vulnerabilities early.\n\nThere are several key dimensions to consider. First is switching cost depth: how painful is it for a customer to leave? [Lattice Labs](companies/lattice-labs-89) exemplifies this well—their integration into customer workflows creates genuine lock-in that compounds over time. Second is value accrual: does the product become more valuable as usage increases? Network effects and data moats fall into this category. Third is contractual structure: multi-year commitments with strong renewal economics signal durability, while month-to-month arrangements do not.\n\n[Beta Labs](companies/beta-labs-51) presents an interesting case study in building durable revenue from day one. Their enterprise focus meant longer sales cycles initially, but the contracts they've secured have 3-year terms with built-in expansion clauses. Compare this to companies optimizing purely for speed-to-revenue who often end up with a customer base that churns aggressivley.\n\nThe framework also helps identify when apparent durability is actually illusory. Vendor lock-in through proprietary formats, for instance, creates short-term stickiness but breeds customer resentment and eventual displacement. True durability comes from continued value delivery, not from trapping customers.\n\n[Acme](companies/acme-0) has been thinking about revenue durability as they scale their platform. Their approach focuses on becoming infrastructure—the kind of service that customers build their own products on top of. This creates natural durability because migration costs scale with customer success.\n\nFor early-stage companies, optimizing for durability might mean slower initial growth but stronger long-term compounding. The tradeoff isn't always worth it, but founders should at least understand what they're trading away when they chase speed over defensibility.",
-  "timeline": "- **2021-06-15** | First internal memo on 'revenue quality vs quantity' circulated among the partnership, laying groundwork for durability framework\n- **2022-02-08** | Concept formally named 'revenue durability' during portfolio review of [Acme](companies/acme-0)'s Series A metrics\n- **2022-09-22** | Framework presented at LP meeting as core evaluation criteria for new investments\n- **2023-03-14** | [Beta Labs](companies/beta-labs-51) cited as exemplar case in durability-first company building during founders dinner\n- **2023-08-30** | Published internal guide: 'Assessing Revenue Durability in Enterprise SaaS'\n- **2024-01-17** | Workshop held with [Lattice Labs](companies/lattice-labs-89) team on strengthening switching costs and contract structure\n- **2024-07-09** | Revenue durability score added to standard deal memo template\n- **2025-02-21** | Framework adapted for AI-native business models after observing new failure modes in portfolio\n- **2025-11-03** | External blog post on durability concepts reaches 50k views, generates significant inbound founder interest",
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/revenue-durability",
-    "name": "revenue durability",
-    "description": "revenue durability as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/beta-labs-51",
-      "companies/acme-0",
-      "companies/lattice-labs-89"
-    ],
-    "related_people": [
-      "people/vera-rodriguez-171",
-      "people/chris-nakamura-40"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__second-time-founder.json b/eval/data/world-v1/concepts__second-time-founder.json
deleted file mode 100644
index 19494fd29..000000000
--- a/eval/data/world-v1/concepts__second-time-founder.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "concepts/second-time-founder",
-  "type": "concept",
-  "title": "Second-Time Founder",
-  "compiled_truth": "The second-time founder occupies a unique position in the startup ecosystem. They carry scar tissue from previous failures—or the confidence from previous exits—and this fundamentally shapes how they approach company building. Unlike first-time founders who often learn by painful trial and error, second-timers have pattern-matched their way through at least one full cycle. They know what a bad hire feels like before the damage spreads. They understand that runway is a countdown clock, not a safety net.\n\nAs a strategic frame, the second-time founder lens helps us evaluate both risk and upside differently. These founders typically raise faster and at better terms because investors recognize the de-risking that comes with experience. They're less likely to make the classic mistakes around premature scaling or over-hiring. But there's a flip side: some second-timers become overly cautious, optimizing for what worked last time rather than what the current moment demands. The best ones treat their previous company as a single data point, not a playbook.\n\n[Epsilon Labs](companies/epsilon-labs-54) exemplifies this pattern well. The founding team includes operators who previously built and sold a developer tools company, and you can see that experience reflected in their measured approach to growth and their focus on unit economics from day one. They didn't need to learn the hard way that enterprise sales cycles are long—they built that into their financial model from the start.\n\nContrast this with [Ranger](companies/ranger-22), where the CEO is building her second company after a difficult shutdown of her first. The lessons there were more about team dynamics and cofounder alignment than go-to-market strategy. She's spoken publicly about how she now spends far more time on culture and communication than she did before.\n\n[Delta](companies/delta-3) represents an interesting edge case—a second-time founder who's building in an entirely different domain than their first company. The transferrable skills are more about operational rigor than domain expertise. Whether this helps or hurts remains to be seen, but early signs suggest the founder's network and fundraising ability have accelerated their timeline considerably.\n\nThe second-time founder frame isn't about hero worship. It's about understanding which lessons transfer and which don't.",
-  "timeline": "- **2021-03-15** | Internal thesis doc first circulated on evaluating repeat founders differently in diligence\n- **2021-09-22** | Met with [Ranger](companies/ranger-22) CEO to discuss her transition from first to second company\n- **2022-04-08** | Published brief analysis on second-time founder performance in Series A outcomes\n- **2022-11-30** | [Epsilon Labs](companies/epsilon-labs-54) founders referenced as case study in LP update\n- **2023-06-14** | Panel discussion at founder summit on \"What You Actually Learn the First Time\"\n- **2023-12-01** | Updated framework to include domain-switching founders like [Delta](companies/delta-3) team\n- **2024-05-19** | Noted pattern: second-timers shipping MVPs 40% faster on average in our portfolio\n- **2024-10-03** | Workshop session with three second-time founders on common pitfalls to avoid\n- **2025-02-28** | Revised thesis to account for AI-native second-time founders emerging from big tech",
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/second-time-founder",
-    "name": "second-time founder",
-    "description": "second-time founder as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/epsilon-labs-54",
-      "companies/ranger-22",
-      "companies/delta-3"
-    ],
-    "related_people": [
-      "people/ian-kapoor-162",
-      "people/tina-moore-191"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__top-down-sales.json b/eval/data/world-v1/concepts__top-down-sales.json
deleted file mode 100644
index 41e191d11..000000000
--- a/eval/data/world-v1/concepts__top-down-sales.json
+++ /dev/null
@@ -1,32 +0,0 @@
-{
-  "slug": "concepts/top-down-sales",
-  "type": "concept",
-  "title": "Top-Down Sales",
-  "compiled_truth": "Top-down sales is a go-to-market strategy where you sell to the executive suite first and let adoption trickle down through the organization. It's the opposite of product-led growth or bottoms-up motions where individual contributors discover and champion a tool. The thesis here is simple: if you can get the CEO or CIO to mandate a solution, you skip the painful process of convincing every team lead and middle manager along the way.\n\nThis approach works best when the product touches multiple departments, requires significant behavior change, or involves substantial budget allocation. Enterprise infrastructure, compliance tools, and strategic platforms tend to fit this mold. You're not selling features—you're selling vision, ROI narratives, and risk mitigation. The sales cycle is longer but deal sizes are dramatically larger.\n\n[Vox](companies/vox-25) exemplifies this strategy in the AI communications space. They deliberately target Chief Revenue Officers and CEOs rather than individual sales reps. Their pitch decks focus on organizational transformation rather than feature comparisons. The result is six-figure contracts instead of seat-based pricing wars. [Wisp](companies/wisp-26) took a similar approach in healthcare, going directly to hospital system executives rather than trying to win over individual clinicians one by one.\n\nThe downsides are real though. Top-down sales requires a more expensive go-to-market team—you need people who can speak boardroom language and navigate complex procurement. Theres also the risk of mandated tools becoming shelfware if you dont eventually win over the actual users. Some founders call this the \"deployment trap.\"\n\n[Vector Labs](companies/vector-labs-56) represents an interesting hybrid model. They started with a bottoms-up motion among ML engineers but pivoted to top-down when they realized enterprise AI governance decisions happen at the C-suite level. Their expansion revenue actually improved after the shift because executives would expand contracts proactively rather than waiting for organic seat growth.\n\nThe strategic frame for company building is this: choose your sales motion based on who controls budget and who controls adoption. When those are the same person, top-down is often the right call. When they're diferent people, you need to carefully sequence how you win both.",
-  "timeline": [
-    "- **2021-08-15** | Internal memo drafted on top-down vs bottoms-up GTM strategies for B2B SaaS portfolio companies",
-    "- **2022-03-22** | [Vox](companies/vox-25) shifts from SMB to enterprise, adopts pure top-down motion targeting CROs",
-    "- **2022-09-10** | Workshop held on executive selling techniques, attended by 6 portfolio company founders",
-    "- **2023-02-14** | [Wisp](companies/wisp-26) closes first $400k health system contract using top-down approach",
-    "- **2023-07-28** | Published internal analysis showing 3.2x higher ACV for companies with top-down sales vs PLG in similar verticals",
-    "- **2024-01-19** | [Vector Labs](companies/vector-labs-56) completes GTM pivot from bottoms-up to executive-led sales",
-    "- **2024-06-05** | Partner discussion on when top-down fails: identified 'deployment trap' as key risk factor",
-    "- **2024-11-30** | Cross-portfolio benchmarking shows top-down companies averaging 18-month sales cycles but 5x deal sizes",
-    "- **2025-04-12** | Added top-down sales fit as explicit diligence criteria for Series A infrastructure deals"
-  ],
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/top-down-sales",
-    "name": "top-down sales",
-    "description": "top-down sales as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/vox-25",
-      "companies/wisp-26",
-      "companies/vector-labs-56"
-    ],
-    "related_people": [
-      "people/grace-martinez-109",
-      "people/grace-singh-197"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__unit-economics.json b/eval/data/world-v1/concepts__unit-economics.json
deleted file mode 100644
index f1915b9d1..000000000
--- a/eval/data/world-v1/concepts__unit-economics.json
+++ /dev/null
@@ -1,32 +0,0 @@
-{
-  "slug": "concepts/unit-economics",
-  "type": "concept",
-  "title": "Unit Economics",
-  "compiled_truth": "Unit economics represents the fundamental building block of sustainable company building—the revenues and costs associated with a single unit of whatever you're selling. Whether that unit is a customer, a subscription, a transaction, or a widget, understanding the economics at this granular level determines whether scaling up will create value or just multiply losses.\n\nThe core insight is deceptively simple: if you lose money on each unit, you cannot make it up in volume. Yet founders regularly fall into this trap, hypnotized by top-line growth while ignoring the contribution margin underneath. Good unit economics means your customer lifetime value (LTV) substntially exceeds your customer acquisition cost (CAC)—typically by 3x or more for healthy SaaS businesses.\n\n[Wisp](companies/wisp-26) exemplifies strong unit economics thinking in healthcare. By focusing on specific conditions with predictable treatment protocols, they've built a model where each patient acquisition translates to a known revenue stream with manageable fulfillment costs. Their CAC payback sits under six months, giving them confidence to pour fuel on growth. Contrast this with generalist telehealth plays that struggled with unpredictable visit patterns and thin margins.\n\nFor AI infrastructure companies like [Vellum Labs](companies/vellum-labs-99), unit economics takes on a differnet character. The unit isn't always a customer—sometimes it's an API call, a model evaluation, or compute time. Vellum has been thoughtful about pricing architecture that scales with customer value creation rather than just usage, ensuring margins expand rather than compress as customers grow.\n\n[Helix](companies/helix-9) demonstrates how unit economics evolve through company stages. Early on, you might intentionally run negative unit economics to prove product-market fit and capture market share. But there needs to be a credible path to positive economics—either through price increases, cost reductions via scale, or expansion revenue from existing customers.\n\nThe strategic frame matters most when making scaling decisions. Pouring capital into a business with broken unit economics just accelerates the burn. But once you've proven healthy unit economics, aggressive investment becomes rational—each incremental customer genuinely creates value. This is why savvy investors scrutinize cohort-level economics obsesively. The aggregate numbers can hide all manner of sins; the cohort economics reveal truth.",
-  "timeline": [
-    "- **2021-06-14** | Internal memo circulated on unit economics frameworks for early-stage evaluation criteria",
-    "- **2022-02-08** | Workshop session with [Wisp](companies/wisp-26) founders on CAC payback modeling for telehealth",
-    "- **2022-09-22** | Published thinking piece on why AI companies need different unit economics benchmarks than traditional SaaS",
-    "- **2023-03-15** | Deep dive with [Vellum Labs](companies/vellum-labs-99) on usage-based pricing and margin dynamics",
-    "- **2023-08-30** | Portfolio-wide analysis of contribution margins across 40+ companies completed",
-    "- **2024-01-12** | Cohort economics review session with [Helix](companies/helix-9) showing improving payback periods",
-    "- **2024-07-19** | Developed updated LTV/CAC benchmarks reflecting 2024 market conditions and capital costs",
-    "- **2025-02-03** | Partner meeting discussion on unit economics red flags in current pipeline deals",
-    "- **2025-11-08** | Presented unit economics framework at annual founder summit, 200+ attendees"
-  ],
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/unit-economics",
-    "name": "unit economics",
-    "description": "unit economics as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/wisp-26",
-      "companies/vellum-labs-99",
-      "companies/helix-9"
-    ],
-    "related_people": [
-      "people/david-singh-73",
-      "people/vera-wilson-25"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__usage-based-pricing.json b/eval/data/world-v1/concepts__usage-based-pricing.json
deleted file mode 100644
index 7d10ed18c..000000000
--- a/eval/data/world-v1/concepts__usage-based-pricing.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "concepts/usage-based-pricing",
-  "type": "concept",
-  "title": "Usage-Based Pricing",
-  "compiled_truth": "Usage-based pricing represents a fundamental shift in how software companies capture value. Rather than charging flat subscriptions, companies meter actual consumption—API calls, compute hours, seats active, messages sent. The model aligns vendor incentives with customer outcomes in ways that traditional SaaS never could.\n\nThe strategic implications run deeper than billing mechanics. Usage-based pricing forces companies to obsess over product value because revenue directly correlates with how much customers actually use the thing. There's nowhere to hide. A customer paying per API call will churn the moment they find a cheaper or better alternative. This creates a natural selection pressure toward building genuinely useful software.\n\n[Forge Labs](companies/forge-labs-69) has become something of a poster child for this approach in the AI tooling space. Their consumption model for inference credits means enterprise customers can start small and scale spend as they prove out use cases internally. The land-and-expand motion is baked into the pricing itself. Similarly, [Cipher](companies/cipher-13) adopted usage-based billing for their security monitoring platform—charging per event ingested rather than per seat. This let them penetrate mid-market accounts that would have balked at enterprise contracts.\n\nThe model isn't without tradeoffs though. Revenue predictibility suffers. Finance teams hate forecasting when usage can swing 40% month to month. Public market investors have historically punished usage-based companies during downturns when customers can instantly reduce spend. Twilio learned this the hard way.\n\nThere's also the cold start problem. Usage-based pricing works beautifully when customers are already using your product heavily. But convincing someone to adopt something new when they can't predict costs? Thats harder. [Lattice Labs](companies/lattice-labs-89) tried pure usage-based early on and found enterprise procurement teams rejecting deals because they couldn't get budget approval without a fixed number.\n\nThe emerging consensus seems to be hybrid models—some base platform fee plus usage components. This gives customers cost predictability while preserving the alignment benefits. The best implementations make the usage component feel like upside rather than risk. You're not paying more because we're gouging you; you're paying more because the product is working.",
-  "timeline": "- **2021-03-15** | Wrote initial memo on consumption pricing after Snowflake S-1 deep dive\n- **2021-09-22** | Discussed usage-based GTM implications with [Forge Labs](companies/forge-labs-69) founders during seed diligence\n- **2022-04-10** | Hosted internal session on metering infrastructure requirements for portfolio companies\n- **2022-11-03** | [Cipher](companies/cipher-13) presented case study on transitioning from seat-based to event-based pricing\n- **2023-02-28** | Published framework comparing pure usage vs hybrid models across 12 portfolio companies\n- **2023-08-14** | Debated revenue predictability concerns with [Lattice Labs](companies/lattice-labs-89) CFO ahead of Series B\n- **2024-01-19** | Noted market sentiment shift—investors increasingly skeptical of pure consumption models post-2023 correction\n- **2024-06-07** | Updated thesis to emphasize hybrid approaches after reviewing churn data across cohorts\n- **2025-02-11** | AI inference pricing emerging as dominant use case; revisited [Forge Labs](companies/forge-labs-69) unit economics",
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/usage-based-pricing",
-    "name": "usage-based pricing",
-    "description": "usage-based pricing as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/forge-labs-69",
-      "companies/cipher-13",
-      "companies/lattice-labs-89"
-    ],
-    "related_people": [
-      "people/chris-miller-101",
-      "people/yara-smith-30"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__vertical-saas.json b/eval/data/world-v1/concepts__vertical-saas.json
deleted file mode 100644
index 6b82abc7e..000000000
--- a/eval/data/world-v1/concepts__vertical-saas.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "concepts/vertical-saas",
-  "type": "concept",
-  "title": "Vertical SaaS",
-  "compiled_truth": "Vertical SaaS represents a fundamental shift in how we think about software company building. Rather than creating horizontal tools that serve everyone poorly, vertical SaaS companies go deep into a specific industry—learning its workflows, regulatory quirks, and tribal knowledge—then building software that feels like it was made by insiders. The thesis is simple: specificity wins.\n\nThe strategic frame here matters more than the label. When you build for a vertical, you're not just picking a market segment, you're choosing a compounding advantage. Every customer conversation teaches you something competitors in adjacent verticals will never learn. Your product roadmap gets shaped by real practioners, not abstract user personas. [Spire Labs](companies/spire-labs-96) exemplifies this approach in their space—they didn't try to be everything to everyone, they picked their lane and dominated it.\n\nWhat makes vertical SaaS defensible isn't just domain expertise, its the network effects that emerge from density. When you own 40% of dental practices in a region, you can offer inter-practice communication, benchmarking data, supplier negotiations. Horizontal players can't touch this. The playbook compounds.\n\nThere's a tension worth acknowledging tho. Going vertical means accepting a smaller TAM upfront. Investors sometimes struggle with this—they want to see massive addressable markets from day one. But the best vertical SaaS founders understand that owning 80% of a $500M market beats owning 2% of a $10B market. [Gamma](companies/gamma-2) has navigated this tension well, proving you can build substantial outcomes in focused verticals.\n\nThe emergence of AI is reshaping vertical SaaS dynamics. Domain-specific training data becomes a moat. [Sentinel Labs](companies/sentinel-labs-73) shows how vertical focus plus AI creates compounding advantages that horizontal competitors simply cannot replicate. When your models are trained on industry-specific edge cases, you're not just better—you're categorically different.\n\nVertical SaaS isn't a business model, its a strategic posture. It says: we will know this industry better than anyone, and that knowledge will compound into product, distribution, and pricing advantages that make us unassailable.",
-  "timeline": "- **2021-03-15** | First internal memo outlining vertical SaaS as core investment thesis for the fund\n- **2021-09-22** | Published \"Why Vertical Wins\" essay, gained traction in founder circles\n- **2022-04-10** | [Spire Labs](companies/spire-labs-96) deal closed, first major vertical SaaS bet in portfolio\n- **2022-11-03** | Hosted vertical SaaS founder dinner with 12 portfolio companies in SF\n- **2023-02-28** | [Gamma](companies/gamma-2) Series A investment, thesis validation in new vertical\n- **2023-08-14** | Debate with partner on TAM limitations of vertical approach—ultimately reinforced conviction\n- **2024-01-19** | [Sentinel Labs](companies/sentinel-labs-73) seed investment, first AI-native vertical SaaS position\n- **2024-06-07** | Keynote at SaaStr on vertical SaaS defensibility in the AI era\n- **2025-02-11** | Internal review shows vertical SaaS portfolio outperforming horizontal bets by 2.3x on markups\n- **2025-09-30** | Started writing book chapter on vertical SaaS for upcoming investing anthology",
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/vertical-saas",
-    "name": "vertical SaaS",
-    "description": "vertical SaaS as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/spire-labs-96",
-      "companies/gamma-2",
-      "companies/sentinel-labs-73"
-    ],
-    "related_people": [
-      "people/grace-miller-122",
-      "people/kevin-taylor-102"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/concepts__wallet-share.json b/eval/data/world-v1/concepts__wallet-share.json
deleted file mode 100644
index 708e93be6..000000000
--- a/eval/data/world-v1/concepts__wallet-share.json
+++ /dev/null
@@ -1,31 +0,0 @@
-{
-  "slug": "concepts/wallet-share",
-  "type": "concept",
-  "title": "Wallet Share",
-  "compiled_truth": "Wallet share is a strategic lens for evaluating how much of a customer's total spend in a category a company can capture over time. Unlike market share, which measures competitive position across all customers, wallet share focuses on depth of relationship with existing customers. The question shifts from 'how many customers do we have' to 'how much of each customer's relevant budget flows through us.'\n\nThis framing becomes particularly powerful in enterprise software and infrastructure plays. [Vellum](companies/vellum-49) exemplifies the wallet share expansion playbook—starting with a narrow wedge (LLM orchestration) and systematically expanding into adjacent budget pools: evaluation, monitoring, prompt management. Each expansion captures more of the AI infrastructure dollar that was previously going to point solutions or internal tooling. The genius is that switching costs compound with each additional surface area captured.\n\n[Prism Labs](companies/prism-labs-93) represents a different wallet share strategy, one built around becoming the system of record for a critical workflow. When you own the canonical data, expanding wallet share becomes almost gravitational—customers pull you into adjacent use cases rather than you having to push. This is the Figma playbook, the Stripe playbook.\n\nThe wallet share frame also helps identify when a market is ripe for consolidation versus fragmentation. High wallet share potential suggests winner-take-most dynamics within accounts, even if the broader market remains competitive. [Nexus Labs](companies/nexus-labs-91) is betting on this with their full-stack approach to simulation—capturing compute, modeling, and visualization spend that currently fragments across vendors.\n\nPractically, wallet share thinking changes how you prioritize product roadmap. Instead of chasing new logos, you invest heavily in platform capabilities that let existing customers do more. Customer success becomes less about retention and more about expansion revenue. The north star metric shifts from CAC payback to net revenue retention.\n\nCritics argue wallet share obsession leads to bloated products and loss of focus. There's truth to this—plenty of companies have destroyed their core value prop chasing adjacencies. The art is identifying expansion paths that genuinely compound customer value versus those that just extract more dollars. The best wallet share plays create genuine lock-in through integrated workflows, not artificial switching costs.",
-  "timeline": [
-    "- **2021-06-14** | Internal thesis doc first uses 'wallet share' as primary evaluation framework for infra investments",
-    "- **2022-03-22** | [Vellum](companies/vellum-49) seed deck explicitly references wallet share expansion as core strategy",
-    "- **2022-11-08** | Partner meeting debates wallet share vs land-and-expand—concludes they're complementary frames",
-    "- **2023-04-17** | [Prism Labs](companies/prism-labs-93) Series A memo cites wallet share potential as key conviction driver",
-    "- **2023-09-30** | Published internal guide on evaluating wallet share potential in vertical SaaS",
-    "- **2024-02-12** | [Nexus Labs](companies/nexus-labs-91) pitch explicitly maps their product roadmap to customer wallet share capture",
-    "- **2024-07-25** | LP meeting uses wallet share framework to explain portfolio construction thesis",
-    "- **2025-01-14** | Noted that highest-performing portfolio companies all exhibit >140% NRR, validating wallet share focus"
-  ],
-  "_facts": {
-    "type": "concept",
-    "slug": "concepts/wallet-share",
-    "name": "wallet share",
-    "description": "wallet share as a strategic frame for thinking about company building.",
-    "related_companies": [
-      "companies/prism-labs-93",
-      "companies/vellum-49",
-      "companies/nexus-labs-91"
-    ],
-    "related_people": [
-      "people/eric-miller-35",
-      "people/frank-moore-167"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__board-acme-2025-q1-0.json b/eval/data/world-v1/meetings__board-acme-2025-q1-0.json
deleted file mode 100644
index 1f33fdf0c..000000000
--- a/eval/data/world-v1/meetings__board-acme-2025-q1-0.json
+++ /dev/null
@@ -1,20 +0,0 @@
-{
-  "slug": "meetings/board-acme-2025-q1-0",
-  "type": "meeting",
-  "title": "Acme Board Meeting Q1 2025",
-  "compiled_truth": "The Q1 2025 board meeting for [Acme](companies/acme-0) convened on January 15th with a focused agenda centered on scaling manufacturing operations and Series B preparation. [Mia Brown](people/mia-brown-0) led the session, opening with a comprehensive review of Q4 2024 performance metrics before diving into strategic planning for the year ahead.\n\nAttendance was smaller than usual—just three board members present. [Chris Jackson](people/chris-jackson-91) dialed in from Singapore where he'd been meeting with potential APAC distribution partners. [Ian Anderson](people/ian-anderson-105) attended in person at Acme's headquarters, bringing detailed financial projections he'd been working on with the finance team over the holidays. Mia facilitated from the head of the table, her laptop open to the slide deck she'd finalized the night before.\n\nKey discussion centered on the robotics company's path to profitability. Ian raised concerns about burn rate, noting that runway would hit critical levels by Q3 without additional funding or revenue acceleration. Chris pushed back somewhat, arguing that aggressive growth spending was justified given the market opportunity. The board ultimately agreed to pursue a Series B round targeting $40M, with outreach to begin in Febuary.\n\nManufacturing capacity dominated the second half of the meeting. Acme's flagship warehouse automation robots had seen demand outpace production for three consecutive quarters. The team debated whether to expand the existing facility in Austin or establish a second production line elsewhere. No final decision was reached, but Ian volunteered to conduct a cost-benefit analysis comparing domestic vs. offshore options.\n\nOther topics included a brief update on the IP portfolio (two new patents pending), discussion of a potential strategic partnership with a major logistics provider, and review of the executive compensation structure. [Chris Jackson](people/chris-jackson-91) abstained from the comp discussion due to conflict of interest concerns.\n\nThe meeting adjourned after roughly two hours with action items assigned and a follow-up scheduled for late March. Overall tone was cautiously optimistic—the company faces real challenges but the board seems aligned on the path forward.",
-  "timeline": "- **2024-10-18** | Q4 board meeting held; initial Series B discussions begin\n- **2024-11-22** | [Ian Anderson](people/ian-anderson-105) joins Acme board as independent director\n- **2024-12-10** | Acme closes largest single customer contract to date\n- **2025-01-08** | [Mia Brown](people/mia-brown-0) circulates pre-read materials and draft agenda\n- **2025-01-15** | Q1 board meeting convenes with three attendees present\n- **2025-01-16** | Meeting minutes distributed to full board for review\n- **2025-02-01** | Series B investor outreach scheduled to begin\n- **2025-03-28** | Follow-up board session planned to review manufacturing analysis",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/board-acme-2025-q1-0",
-    "name": "Acme Board Meeting Q1",
-    "meeting_type": "board_meeting",
-    "date": "2025-01-15",
-    "attendees": [
-      "people/mia-brown-0",
-      "people/chris-jackson-91",
-      "people/ian-anderson-105"
-    ],
-    "topic_company": "companies/acme-0"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__board-beta-2025-q2-1.json b/eval/data/world-v1/meetings__board-beta-2025-q2-1.json
deleted file mode 100644
index 171e7d092..000000000
--- a/eval/data/world-v1/meetings__board-beta-2025-q2-1.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "meetings/board-beta-2025-q2-1",
-  "type": "meeting",
-  "title": "Beta Board Meeting Q2 2025",
-  "compiled_truth": "The Q2 2025 board meeting for [Beta](companies/beta-1) convened on April 15th with a notably slim attendee list. [Victor Taylor](people/victor-taylor-1) was present, though the sparse turnout raised some eyebrows given the agenda weight. The meeting ran approximately two hours and covered ground ranging from financial performance to product roadmap adjustments.\n\nPrimary discussion centered on Beta's cybersecurity platform expansion into the mid-market segment. Victor pushed hard on unit economics, questioning whether the company's enterprise-focused sales motion could translate effectively to smaller deal sizes. The sales team had been experimenting with a product-led growth motion for three months, results were mixed at best. CAC payback periods stretched longer than projected, though retention metrics showed promise once customers hit the 90-day mark.\n\nRevenue came in at $4.2M for Q1, slightly under forecast but within acceptable variance. The miss was attributed to two delayed enterprise deals that slipped into Q2—both now signed. ARR stood at $18.1M with net revenue retention holding steady at 112%. Victor noted that while these numbers were solid, the company needed to accelerate if it wanted to hit the $30M target by year end.\n\nA significant portion of the meeting addressed competitive dynamics. Two well-funded startups had emerged in Beta's core market segment, and a larger player had begun bundling similar functionality into their platform. The board discussed potential responses including feature acceleration, strategic partnerships, and pricing adjustments. No final decisions were made but the team committed to presenting options at the next session.\n\nHeadcount planning also came up. Beta currently sits at 67 employees and had budget for 15 additional hires in 2025. Engineering and sales were the priority functions. Victor recommended front-loading engineering hires to ensure the product roadmap stayed on track, even if it meant delaying some sales expansion. The meeting adjourned with action items assigned and a follow-up scheduled for late May to review Q2 progress.",
-  "timeline": "- **2024-01-22** | Beta closes Series A, [Victor Taylor](people/victor-taylor-1) joins board\n- **2024-04-10** | Q1 2024 board meeting held, first official session with full board\n- **2024-07-18** | Mid-year strategy review, pivot toward mid-market approved\n- **2024-10-09** | Q3 board meeting discusses competitive landscape shifts\n- **2025-01-14** | Q4 2024 board review, annual planning session completed\n- **2025-02-28** | [Beta](companies/beta-1) launches product-led growth experiment\n- **2025-04-15** | Q2 board meeting convenes with limited attendance\n- **2025-05-29** | Follow-up session scheduled to review Q2 trajectory",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/board-beta-2025-q2-1",
-    "name": "Beta Board Meeting Q2",
-    "meeting_type": "board_meeting",
-    "date": "2025-04-15",
-    "attendees": [
-      "people/victor-taylor-1"
-    ],
-    "topic_company": "companies/beta-1"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__board-delta-2025-q4-3.json b/eval/data/world-v1/meetings__board-delta-2025-q4-3.json
deleted file mode 100644
index 2c7846e00..000000000
--- a/eval/data/world-v1/meetings__board-delta-2025-q4-3.json
+++ /dev/null
@@ -1,21 +0,0 @@
-{
-  "slug": "meetings/board-delta-2025-q4-3",
-  "type": "meeting",
-  "title": "Delta Board Meeting Q4 2025",
-  "compiled_truth": "Quarterly board meeting for [Delta](companies/delta-3) held on October 15, 2025, convening the full board to review Q4 performance projections and strategic initiatives heading into 2026. The meeting ran approximately three hours with all members present in person at Delta's South San Francisco headquarters.\n\n[Victor Wilson](people/victor-wilson-3) opened the session with a CEO update covering the Phase 2 clinical trial results for DLT-4401, Delta's lead oncology candidate. Results showed statistically significant tumor reduction in 67% of patients, exceeding the 55% threshold the board had set as a go/no-go criteria for Phase 3 planning. Victor noted that the data package would be ready for FDA pre-Phase 3 meeting by Q1 2026.\n\n[David Zhang](people/david-zhang-83) presented the financial overview as board chair. Cash runway currently sits at 18 months with $47M remaining from the Series C. David raised concerns about the burn rate acceleration—R&D spend increased 23% quarter-over-quarter due to expanded trial sites. He recommended the board begin socializing a Series D in Q1, targeting $80-100M to fund Phase 3 and commercial preparation.\n\n[Rachel Brown](people/rachel-brown-95) led discussion on the competitive landscape. Two competitors have entered Phase 2 with similar mechanisms, though Rachel emphasized Delta's first-mover advantage and superior safety profile. She pushed for accelerating the timeline on the companion diagnostic partnership, noting that Roche had expresed interest in co-development.\n\n[David Brown](people/david-brown-187) raised governance matters, including the upcoming CFO search. The current interim CFO arrangement expires in December, and David recomended engaging an executive search firm immediately. He also flagged that two independent board seats should be filled before any Series D to strengthen the cap table story for institutional investors.\n\nKey resolutions passed: approval of Phase 3 budget planning, authorization to engage Cooley LLP for Series D preparation, and greenlight on the Roche diagnostic partnership term sheet negotiation. Next board meeting scheduled for January 2026.",
-  "timeline": "- **2023-06-12** | Delta completes Series B, [Victor Wilson](people/victor-wilson-3) joins as CEO\n- **2024-02-20** | First patient dosed in Phase 2 trial for DLT-4401\n- **2024-09-15** | [David Zhang](people/david-zhang-83) appointed board chair following founder transition\n- **2025-01-22** | Series C closes at $52M led by OrbiMed\n- **2025-04-08** | Q1 board meeting approves expanded trial sites in Europe\n- **2025-07-14** | Interim CFO arrangement established after departure of finance lead\n- **2025-08-30** | Phase 2 interim data readout shows promising efficacy signals\n- **2025-10-15** | Q4 board meeting held, Phase 3 planning approved\n- **2025-12-01** | CFO search committee formed with [David Brown](people/david-brown-187) as lead\n- **2026-01-28** | Next quarterly board meeting scheduled",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/board-delta-2025-q4-3",
-    "name": "Delta Board Meeting Q4",
-    "meeting_type": "board_meeting",
-    "date": "2025-10-15",
-    "attendees": [
-      "people/victor-wilson-3",
-      "people/david-zhang-83",
-      "people/rachel-brown-95",
-      "people/david-brown-187"
-    ],
-    "topic_company": "companies/delta-3"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__board-epsilon-2025-q1-4.json b/eval/data/world-v1/meetings__board-epsilon-2025-q1-4.json
deleted file mode 100644
index 252d1d762..000000000
--- a/eval/data/world-v1/meetings__board-epsilon-2025-q1-4.json
+++ /dev/null
@@ -1,21 +0,0 @@
-{
-  "slug": "meetings/board-epsilon-2025-q1-4",
-  "type": "meeting",
-  "title": "Epsilon Board Meeting Q1 2025",
-  "compiled_truth": "The Q1 board meeting for [Epsilon](companies/epsilon-4) convened on January 15th, 2025, bringing together key stakeholders to review the cybersecurity firm's performance and strategic direction. The session ran approximately three hours and covered financials, product roadmap, and go-to-market expansion plans.\n\n[Paul Rodriguez](people/paul-rodriguez-4) opened with a CEO update, noting that Epsilon closed 2024 with ARR of $18.2M, representing 84% year-over-year growth. He emphasized that the enterprise segment drove most of this momentum, with average deal sizes increasing to $340K from $215K the prior year. Paul acknowledged some challenges in the mid-market segment where competition from CrowdStrike and SentinelOne has intensified. He's proposing a dedicated SMB product tier to address this gap.\n\n[Sarah Lopez](people/sarah-lopez-84) presented the financial deep-dive, walking through burn rate, runway, and unit economics. Gross margins improved to 72%, up from 68% in Q3. She flagged that sales & marketing spend as a percentage of revenue remains elevated at 58%, though this is expected given the growth phase. Sarah recommened exploring debt financing options to extend runway without further dilution ahead of a potential Series C.\n\nProduct discussion was led by [Sarah Williams](people/sarah-williams-92), who demoed the new threat intelligence dashboard shipping in February. The board was impressed by the real-time correlation engine, though questions arose about infrastructure costs at scale. Sarah W. committed to providing detailed cost modeling by the next board session. She also mentioned early explorations into AI-powered anomaly detection, which generated significant interest.\n\n[Olivia Miller](people/olivia-miller-176) provided investor perspective and pushed hard on competitive positioning. She wants to see clearer differentiation messaging and suggested bringing in an external brand consultant. Olivia also raised concerns about key-person risk given the technical team's concentration around a few senior engineers.\n\nAction items: Paul to finalize SMB product proposal by end of Febuary. Sarah Lopez to present debt financing options at next meeting. Sarah Williams to deliver infrastructure cost projections. Olivia to connect Epsilon with brand consultants from her network.",
-  "timeline": "- **2024-10-08** | [Sarah Lopez](people/sarah-lopez-84) joins Epsilon as CFO, bringing experience from two prior cybersecurity exits\n- **2024-11-15** | Q4 board meeting held; approved 2025 headcount plan for 45 new hires\n- **2024-12-02** | [Epsilon](companies/epsilon-4) closes largest enterprise deal to date with Fortune 500 financial services firm\n- **2025-01-07** | Pre-board strategy session between [Paul Rodriguez](people/paul-rodriguez-4) and [Olivia Miller](people/olivia-miller-176)\n- **2025-01-15** | Q1 board meeting convenes; ARR growth and product roadmap reviewed\n- **2025-01-22** | [Sarah Williams](people/sarah-williams-92) begins infrastructure cost analysis following board request\n- **2025-02-10** | Threat intelligence dashboard v2.0 scheduled for GA release\n- **2025-04-15** | Next board meeting scheduled to review Series C readiness",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/board-epsilon-2025-q1-4",
-    "name": "Epsilon Board Meeting Q1",
-    "meeting_type": "board_meeting",
-    "date": "2025-01-15",
-    "attendees": [
-      "people/paul-rodriguez-4",
-      "people/sarah-lopez-84",
-      "people/sarah-williams-92",
-      "people/olivia-miller-176"
-    ],
-    "topic_company": "companies/epsilon-4"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__board-gamma-2025-q3-2.json b/eval/data/world-v1/meetings__board-gamma-2025-q3-2.json
deleted file mode 100644
index a7f42df12..000000000
--- a/eval/data/world-v1/meetings__board-gamma-2025-q3-2.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/board-gamma-2025-q3-2",
-  "type": "meeting",
-  "title": "Gamma Board Meeting Q3 2025",
-  "compiled_truth": "The Q3 2025 board meeting for [Gamma](companies/gamma-2) convened on July 15th with a notably lean attendee list. [Mark Jones](people/mark-jones-2) chaired the session, joined by [Vera Gonzalez](people/vera-gonzalez-103) who participated remotely from the company's new Austin satellite office. The streamlined attendance reflected Gamma's ongoing effort to make board meetings more efficient and decision-oriented rather than performative.\n\nPrimary discussion centered on Gamma's expansion into embedded lending products. Mark opened with a review of Q2 financials, noting that transaction volume had exceeded projections by roughly 12% but that customer acqusition costs remained stubbornly high in the SMB segment. Vera pushed back on the proposed marketing budget increase, arguing that the unit economics wouldn't justify the spend until the product-led growth motion showed more traction. This led to a somewhat heated but productive debate about channel strategy.\n\nThe fintech's regulatory positioning came up repeatedly. With new federal guidelines on instant payment rails expected in Q4, the board discussed whether Gamma should accelerate its compliance infrastructure buildout or wait for final rule clarity. Vera advocated for the aggressive approach, citing competitive dynamics—several larger players were already positioning themselves. Mark expressed concern about resource allocation given the team was already stretched thin on the core payments roadmap.\n\nOther items covered included a preliminary look at Series C timing. The consensus was to delay formal fundraising discussions until Q4, pending resolution of some technical debt issues that have been affecting platform reliability. There was brief mention of a potential acqui-hire target in the fraud detection space, though details were kept high-level pending further diligence.\n\nAction items assigned: Mark to follow up with legal counsel on the regulatory question. Vera commited to reviewing the revised CAC projections before the August check-in. Meeting adjourned after approximately ninety minutes.",
-  "timeline": "- **2025-07-15** | Q3 board meeting held with [Mark Jones](people/mark-jones-2) and [Vera Gonzalez](people/vera-gonzalez-103) in attendance\n- **2025-07-10** | Board deck circulated ahead of meeting; included updated financial projections\n- **2025-05-22** | [Gamma](companies/gamma-2) hits 50,000 active merchant accounts milestone\n- **2025-04-18** | Q2 board meeting; approved expansion into embedded lending vertical\n- **2025-02-03** | Vera Gonzalez joins Gamma's board as independent director\n- **2024-11-15** | Series B extension closed at $28M valuation\n- **2024-08-20** | Mark Jones appointed board chair following founder transition\n- **2024-03-12** | Gamma launches instant settlement feature for enterprise clients",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/board-gamma-2025-q3-2",
-    "name": "Gamma Board Meeting Q3",
-    "meeting_type": "board_meeting",
-    "date": "2025-07-15",
-    "attendees": [
-      "people/mark-jones-2",
-      "people/vera-gonzalez-103"
-    ],
-    "topic_company": "companies/gamma-2"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__board-helix-2025-q2-9.json b/eval/data/world-v1/meetings__board-helix-2025-q2-9.json
deleted file mode 100644
index 92bf8867e..000000000
--- a/eval/data/world-v1/meetings__board-helix-2025-q2-9.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/board-helix-2025-q2-9",
-  "type": "meeting",
-  "title": "Helix Board Meeting Q2 2025",
-  "compiled_truth": "The Q2 2025 board meeting for [Helix](companies/helix-9) convened on April 15th with a focused agenda around infrastructure scaling and the company's expanding enterprise pipeline. [Rachel Garcia](people/rachel-garcia-9) led the session, presenting updated financials and a revised go-to-market strategy that emphasized verticalized sales motions for financial services and healthcare customers.\n\nXavier opened with a review of Q1 performance metrics. Revenue came in at $4.2M for the quarter, up 34% from Q4, though slightly below the internal stretch target of $4.5M. The miss was attributed to two delayed enterprise deals that slipped into April—both have since closed. Gross margins held steady at 71%, which the board noted as encouraging given the compute-intensive nature of Helix's offering.\n\n[Xavier Patel](people/xavier-patel-183) walked through the product roadmap, highlighting the upcoming launch of Helix Runtime 2.0, scheduled for late May. The new runtime promises 40% latency improvements and native support for multi-model orchestration. Board disucssion centered on whether to accelerate the release given competitive pressure from several well-funded startups in the inference optimization space.\n\nA significant portion of the meeting focused on headcount planning. Rachel proposed adding 8 engineers and 3 enterprise AEs in Q2-Q3, which would bring total headcount to 47. The board approved the plan contingent on maintaining a 24-month runway, currently sitting at 26 months post the Series A close. There was some debate about whether to prioritize ML platform engineers versus infrastructure generalists—Xavier argued for the former, citing customer feedback around model deployment complexity.\n\nOther topics included a preliminary discussion of Series B timing. Consensus was to begin informal conversations with growth-stage investors in Q3, targeting a raise in early 2026 if current trajectories hold. Rachel mentioned inbound interest from two tier-one firms but recommended waiting until the enterprise pipeline matures further. The meeting adjurned at 11:45am PT after brief discussion of board composition and potential independent director candidates.",
-  "timeline": "- **2024-08-12** | Helix closes $18M Series A; [Rachel Garcia](people/rachel-garcia-9) joins board as lead director\n- **2024-10-03** | Q3 board meeting approves expansion into healthcare vertical\n- **2024-11-15** | Xavier Patel presents at AI Infrastructure Summit, announces Helix Runtime beta\n- **2025-01-21** | Q1 board meeting held; company hits $3.1M quarterly revenue\n- **2025-03-08** | Helix signs first $1M+ annual contract with major financial institution\n- **2025-04-15** | Q2 board meeting convenes with [Xavier Patel](people/xavier-patel-183) and Rachel Garcia\n- **2025-05-28** | Helix Runtime 2.0 scheduled for general availability\n- **2025-07-15** | Q3 board meeting planned; Series B prep discussions on agenda",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/board-helix-2025-q2-9",
-    "name": "Helix Board Meeting Q2",
-    "meeting_type": "board_meeting",
-    "date": "2025-04-15",
-    "attendees": [
-      "people/rachel-garcia-9",
-      "people/xavier-patel-183"
-    ],
-    "topic_company": "companies/helix-9"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__board-nimbus-2025-q2-5.json b/eval/data/world-v1/meetings__board-nimbus-2025-q2-5.json
deleted file mode 100644
index 1f232a802..000000000
--- a/eval/data/world-v1/meetings__board-nimbus-2025-q2-5.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/board-nimbus-2025-q2-5",
-  "type": "meeting",
-  "title": "Nimbus Board Meeting Q2 2025",
-  "compiled_truth": "The Q2 2025 board meeting for [Nimbus](companies/nimbus-5) took place on April 15th, with a notably lean attendance roster. [Mia Anderson](people/mia-anderson-5) led the session, walking through the company's progress on their atmospheric water harvesting technology and the commercial pilots currently underway in three drought-affected regions. [Noah Nakamura](people/noah-nakamura-182) joined remotely from Tokyo, where he'd been scouting potential manufacturing partners for the next-gen condensation units.\n\nThe meeting opened with a financial review. Runway sits at approximately 14 months, which is tighter than the board would like given macro headwinds in climate tech funding. Mia presented two scenarios: a bridge round from existing investors or an accelerated push toward revenue through enterprise water-as-a-service contracts. Noah pushed back on the bridge approach, arguing that another small raise would signal weakness to the market. He favored the revenue path, even if it ment delaying the Series B by two quarters.\n\nDiscussion then shifted to the pilot deployments. The Chile installation is performing above expectations—12% more yield than lab projections, which Mia attributed to unexpectedly favorable humidity patterns. The Morocco site has hit regulatory snags around water rights that legal is still working through. Arizona is on track but facing some local community pushback that the team is addressing through town halls.\n\nNoah raised concerns about the competitive landscape. Two well-funded startups out of Israel have announced similar atmospheric capture tech, and there's been some patent overlap that counsel is evaluating. The board agreed to allocate budget for a freedom-to-operate analysis before the next raise.\n\nThe meeting wrapped with a discussion of team expansion. Nimbus needs a VP of Sales to drive enterprise deals, and Mia mentioned she's been talking to a few candidates from the industrial water sector. Noah offered to tap his network for introductions. Action items were assigned with a follow-up sync scheduled for early May.",
-  "timeline": "- **2024-09-10** | [Nimbus](companies/nimbus-5) closes seed extension; [Noah Nakamura](people/noah-nakamura-182) joins board as observer\n- **2024-11-02** | First prototype unit achieves sustained 200L/day output in lab conditions\n- **2025-01-18** | Q1 board meeting; decision made to pursue three international pilot sites\n- **2025-02-14** | Chile pilot installation begins with local utility partnership\n- **2025-03-03** | [Mia Anderson](people/mia-anderson-5) presents at Climate Tech Summit on distributed water infrastructure\n- **2025-03-22** | Morocco pilot delayed due to regulatory review of water extraction permits\n- **2025-04-15** | Q2 board meeting held; runway concerns and competitive threats discussed\n- **2025-05-08** | Follow-up board sync scheduled to review bridge vs. revenue strategy\n- **2025-06-01** | Projected: Arizona pilot reaches full operational capacity",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/board-nimbus-2025-q2-5",
-    "name": "Nimbus Board Meeting Q2",
-    "meeting_type": "board_meeting",
-    "date": "2025-04-15",
-    "attendees": [
-      "people/mia-anderson-5",
-      "people/noah-nakamura-182"
-    ],
-    "topic_company": "companies/nimbus-5"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__board-pulse-2025-q1-8.json b/eval/data/world-v1/meetings__board-pulse-2025-q1-8.json
deleted file mode 100644
index b0976f810..000000000
--- a/eval/data/world-v1/meetings__board-pulse-2025-q1-8.json
+++ /dev/null
@@ -1,20 +0,0 @@
-{
-  "slug": "meetings/board-pulse-2025-q1-8",
-  "type": "meeting",
-  "title": "Pulse Board Meeting Q1 2025",
-  "compiled_truth": "The Q1 2025 board meeting for [Pulse](companies/pulse-8) convened on January 15th with a focused agenda centered on growth metrics and the upcoming Series B timeline. [Yara Johnson](people/yara-johnson-8) led the session as board chair, opening with a review of the company's performance against Q4 targets. The meeting ran approximately two hours, longer than typical due to extended discussion around international expansion.\n\n[Eric Martinez](people/eric-martinez-93) presented the financial overview, highlighting that Pulse had reached $4.2M ARR by end of 2024, representing 140% year-over-year growth. He noted that customer acqusition costs had decreased by 18% thanks to improved organic channels and word-of-mouth referrals from existing school district partners. The burn rate remains manageable at roughly $380K monthly, giving the company approximately 14 months of runway at current spend levels.\n\n[David Kim](people/david-kim-186) provided the product and technical update. Key highlights included the successful launch of the adaptive learning module in November, which has already been adopted by 67% of active users. David mentioned that the engineering team has grown to 12 people and they're on track to ship the parent dashboard feature by end of Q1. There was some tension in the room when Yara pushed back on the mobile app timeline, suggesting it should be accelerated given competitor moves.\n\nStrategic discussion focused heavily on whether to pursue the UK market in 2025 or double down on US penetration. Eric advocated for international expansion, citing inbound interest from three UK academy trusts. Yara expressed caution about spreading too thin before the Series B closes. The board ultimately agreed to conduct a lightweight UK pilot with one partner school while maintaining US focus.\n\nAction items from the meeting: prepare Series B materials for March investor outreach, finalize UK pilot terms by February 15th, and schedule a compensation committee review for the executive team. Next board meeting set for April 9th.",
-  "timeline": "- **2024-10-22** | Q4 board meeting held; approved hiring plan for 5 additional engineers\n- **2024-11-15** | [Pulse](companies/pulse-8) launches adaptive learning module to all customers\n- **2024-12-03** | [Eric Martinez](people/eric-martinez-93) presents preliminary 2024 financials to board via email\n- **2024-12-18** | [David Kim](people/david-kim-186) shares product roadmap draft with board members for review\n- **2025-01-08** | Pre-meeting call between [Yara Johnson](people/yara-johnson-8) and Eric to align on agenda\n- **2025-01-15** | Q1 board meeting convenes; Series B timeline and UK expansion discussed\n- **2025-01-16** | Meeting notes and action items distributed to all attendees\n- **2025-02-15** | Target date for UK pilot partnership terms finalization\n- **2025-04-09** | Next scheduled board meeting",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/board-pulse-2025-q1-8",
-    "name": "Pulse Board Meeting Q1",
-    "meeting_type": "board_meeting",
-    "date": "2025-01-15",
-    "attendees": [
-      "people/yara-johnson-8",
-      "people/eric-martinez-93",
-      "people/david-kim-186"
-    ],
-    "topic_company": "companies/pulse-8"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__board-quantum-2025-q4-7.json b/eval/data/world-v1/meetings__board-quantum-2025-q4-7.json
deleted file mode 100644
index d11cd2f15..000000000
--- a/eval/data/world-v1/meetings__board-quantum-2025-q4-7.json
+++ /dev/null
@@ -1,20 +0,0 @@
-{
-  "slug": "meetings/board-quantum-2025-q4-7",
-  "type": "meeting",
-  "title": "Quantum Board Meeting Q4 2025",
-  "compiled_truth": "The Q4 2025 board meeting for [Quantum](companies/quantum-7) convened on October 15th with a focused agenda centered on year-end positioning and 2026 strategic planning. Three board members were present: [Ulrich Johnson](people/ulrich-johnson-7), [Kate Anderson](people/kate-anderson-107), and [Noah Williams](people/noah-williams-198). The meeting ran aproximately three hours, longer than typical due to extensive discussion around international expansion.\n\nUlrich opened with a review of Q3 financials, noting that Quantum had exceeded revenue targets by 12% while maintaining reasonable burn. The fintech's core payment processing infrastructure showed strong adoption among mid-market enterprise clients, though there was some concern about concentration risk with the top five customers representing nearly 40% of ARR. Kate pushed back on this framing, arguing that deep enterprise relationships were actually a competitive moat rather than a vulnerability.\n\nNoah Williams presented competitive landscape analysis, highlighting three emerging players in the embedded finance space that could threaten Quantum's positioning. The board discussed potential acqui-hire targets to accelerate product development, with two candidates identified for further diligence. There was consensus that the engineering team needed to grow by at least 15 heads in Q1 2026.\n\nSignificant time was spent debating European expansion. [Kate Anderson](people/kate-anderson-107) advocated for a Dublin office to serve as EU headquarters, citing regulatory advantages and talent availability. Ulrich expressed caution about spreading resources too thin before achieving profitibility in North America. The board ultimately approved a preliminary budget for EU market research, with a full proposal expected at the Q1 2026 meeting.\n\nOther items covered: board compensation adjustments, D&O insurance renewal, and preliminary discussions about Series C timing. [Noah Williams](people/noah-williams-198) noted that current market conditions favored waiting until mid-2026 for the next raise, assuming growth metrics held steady. The meeting concluded with agreement to schedule a special session in December focused solely on 2026 OKRs.",
-  "timeline": "- **2025-10-01** | Board materials distributed ahead of Q4 meeting, including updated cap table and financial projections\n- **2025-10-15** | Q4 board meeting held with [Ulrich Johnson](people/ulrich-johnson-7), [Kate Anderson](people/kate-anderson-107), and [Noah Williams](people/noah-williams-198) in attendance\n- **2025-10-16** | Follow-up call between Kate and CEO to discuss EU expansion research scope\n- **2025-10-22** | Board approved preliminary EU market research budget via email resolution\n- **2025-11-05** | Special compensation committee call scheduled to finalize exec team adjustments\n- **2025-12-10** | Planned special board session for 2026 strategic planning and OKR review",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/board-quantum-2025-q4-7",
-    "name": "Quantum Board Meeting Q4",
-    "meeting_type": "board_meeting",
-    "date": "2025-10-15",
-    "attendees": [
-      "people/ulrich-johnson-7",
-      "people/kate-anderson-107",
-      "people/noah-williams-198"
-    ],
-    "topic_company": "companies/quantum-7"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__board-vector-2025-q3-6.json b/eval/data/world-v1/meetings__board-vector-2025-q3-6.json
deleted file mode 100644
index 8943af97f..000000000
--- a/eval/data/world-v1/meetings__board-vector-2025-q3-6.json
+++ /dev/null
@@ -1,21 +0,0 @@
-{
-  "slug": "meetings/board-vector-2025-q3-6",
-  "type": "meeting",
-  "title": "Vector Board Meeting Q3 2025",
-  "compiled_truth": "The Q3 2025 board meeting for [Vector](companies/vector-6) convened on July 15th with a focused agenda around expansion planning and regulatory readiness. [Uma Brown](people/uma-brown-6) opened the session with a review of Q2 performance metrics, noting that the company had exceeded ARR targets by 12% while maintaining strong unit economics across their core health monitoring platform.\n\n[David Zhang](people/david-zhang-83) presented the technical roadmap, emphasizing the team's progress on their FDA 510(k) submission for the new cardiac arrythmia detection module. The board spent considerable time discussing the regulatory timeline, with David noting that pre-submission feedback from the FDA had been encouraging. He mentioned that the clinical validation study had enrolled 2,400 patients across six hospital systems, exceeding the statistical power requirements.\n\n[Vera Gonzalez](people/vera-gonzalez-103) raised questions about the competitive landscape, particularly regarding two well-funded startups that had entered the remote patient monitoring space in Q2. Uma acknowledged the increased competition but argued that Vector's integration partnerships with major EHR systems created meaningful switching costs. The discussion turned to whether Vector should accelerate its enterprise sales hiring or maintain the current measured approach.\n\nFinancial planning dominated the second half of the meeting. [Wendy Wilson](people/wendy-wilson-170) walked through several scenarios for the Series C raise, recommending that Vector begin conversations with growth-stage investors in Q4 rather than waiting until early 2026. The board debated valuation expectations given the cooling health tech market—Wendy suggested targeting a $180-220M range based on comparable transactions.\n\nKey resolutions included approval of a $2.1M budget increase for regulatory affairs, authorization to begin Series C preparation, and agreement to revisit the sales hiring plan at the October board session. Uma committed to providing monthly investor pipeline updates to board members. The meeting concluded with a brief executive session focused on upcoming leadership team expansions.",
-  "timeline": "- **2023-11-08** | Vector closes Series A, board formally constituted with initial member appointments\n- **2024-02-20** | First formal board meeting establishes governance framework and committee structure\n- **2024-06-15** | Q2 board meeting approves FDA regulatory strategy presented by [David Zhang](people/david-zhang-83)\n- **2024-10-10** | Board authorizes Series B fundraising process\n- **2025-01-22** | Special board session to approve Series B term sheet, [Wendy Wilson](people/wendy-wilson-170) joins as board observer\n- **2025-04-15** | Q1 board meeting reviews post-Series B integration and hiring plans\n- **2025-07-15** | Q3 board meeting focuses on Series C planning and FDA submission timeline\n- **2025-10-14** | Scheduled Q4 board meeting to review Series C investor pipeline\n- **2026-01-20** | Anticipated board meeting to approve Series C terms",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/board-vector-2025-q3-6",
-    "name": "Vector Board Meeting Q3",
-    "meeting_type": "board_meeting",
-    "date": "2025-07-15",
-    "attendees": [
-      "people/uma-brown-6",
-      "people/david-zhang-83",
-      "people/vera-gonzalez-103",
-      "people/wendy-wilson-170"
-    ],
-    "topic_company": "companies/vector-6"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__demo-day-2024-01-15-batch-0.json b/eval/data/world-v1/meetings__demo-day-2024-01-15-batch-0.json
deleted file mode 100644
index 2485f8c39..000000000
--- a/eval/data/world-v1/meetings__demo-day-2024-01-15-batch-0.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "meetings/demo-day-2024-01-15-batch-0",
-  "type": "meeting",
-  "title": "Demo Day W24 — Acme",
-  "compiled_truth": "Demo Day for the Winter 2024 batch featured [Acme](companies/acme-0) presenting their robotics platform to a packed room of investors and partners. The company had been building toward this moment for months, refining their pitch and tightening the product demo. [Wendy Hernandez](people/wendy-hernandez-80) led the presentation, walking through Acme's core value proposition: affordable robotic arms for small manufacturing operations that can be programmed by workers without engineering backgrounds.\n\nThe pitch ran about four minutes. Wendy opened with the problem—how most robotics solutions price out 90% of potential customers—then handed off to a live demo showing their flagship arm sorting components on a simulated assembly line. The arm fumbled once during the demo (brief nervous laughter from the audience) but recovered smoothly. Investors seemed engaged, several leaning forward during the pricing slide.\n\n[Mia Brown](people/mia-brown-0) and [Mia Anderson](people/mia-anderson-5) were both in attendance representing diferent funds. Brown asked a pointed question about unit economics during the Q&A portion, specifically around margins at the $15k price point Acme is targeting. Anderson followed up on go-to-market, wondering how they'd reach small manufacturers who don't typically attend trade shows or read industry publications. Both seemed skeptical but interested.\n\n[Mark Thomas](people/mark-thomas-11) provided some color commentary afterward—he'd seen Acme's earlier iterations and noted how much tighter the narrative had become. [Chris Smith](people/chris-smith-110) stuck around for the networking session and was spotted in extended conversation with Wendy near the refreshments table. Could be nothing, could be a seed check brewing.\n\nOverall reception was positive but not euphoric. Acme sits in a crowded space and the differentation story needs more work. The live demo helped enormously—investors trust what they can see working. Next steps likely include follow-up meetings with at least three or four funds from today's attendees. The team seemed energized but exhausted, which tracks for demo day.",
-  "timeline": "- **2023-09-01** | [Acme](companies/acme-0) accepted into W24 batch, begins program\n- **2023-10-15** | Early prototype of robotic arm shown at batch office hours\n- **2023-11-20** | [Wendy Hernandez](people/wendy-hernandez-80) presents initial pitch deck to batch partners for feedback\n- **2023-12-08** | Acme completes first manufacturable unit, begins stress testing\n- **2024-01-10** | Dress rehearsal for demo day, feedback from peers\n- **2024-01-15** | Demo Day W24 presentation delivered to investors and partners\n- **2024-01-17** | Follow-up meeting scheduled with [Mia Anderson](people/mia-anderson-5)\n- **2024-01-22** | [Chris Smith](people/chris-smith-110) visits Acme office for deep-dive session\n- **2024-02-14** | Post demo day investor update sent to all attending funds",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/demo-day-2024-01-15-batch-0",
-    "name": "Demo Day W24",
-    "meeting_type": "demo_day",
-    "date": "2024-01-15",
-    "attendees": [
-      "people/wendy-hernandez-80",
-      "people/mia-brown-0",
-      "people/mia-anderson-5",
-      "people/mark-thomas-11",
-      "people/chris-smith-110"
-    ],
-    "topic_company": "companies/acme-0"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__demo-day-2024-02-16-batch-1.json b/eval/data/world-v1/meetings__demo-day-2024-02-16-batch-1.json
deleted file mode 100644
index 2ffea19e3..000000000
--- a/eval/data/world-v1/meetings__demo-day-2024-02-16-batch-1.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "meetings/demo-day-2024-02-16-batch-1",
-  "type": "meeting",
-  "title": "Demo Day W25 - Beta Cybersecurity Pitch",
-  "compiled_truth": "Demo Day W25 featured [Beta](companies/beta-1) presenting their cybersecurity platform to a panel of investors and partners. The session took place on February 16th, 2024 as part of the winter batch showcase. Beta's presentation focused on their AI-driven threat detection system, which has been gaining traction among mid-market enterprises.\n\n[Carol Jackson](people/carol-jackson-81) led the investor contingent, asking pointed questions about Beta's go-to-market strategy and competitive moat against established players like CrowdStrike. She seemed particularly interested in the company's approach to reducing false positives, which Beta claims to have cut by 73% compared to legacy solutions. [Victor Taylor](people/victor-taylor-1) followed up with questions about the team's technical background and previous exits.\n\nThe Beta founding team handled objections well. Their CTO demonstrated the platform live, showing real-time threat correlation across a simulated enterprise network. The demo hit a minor snag when the dashboard froze momentarily, but they recoverd smoothly. [Uma Brown](people/uma-brown-6) noted afterward that the product clearly had strong engineering behind it despite the hiccup.\n\n[Henry Johnson](people/henry-johnson-12) raised concerns about Beta's burn rate relative to their current ARR of $890K. The founders acknowledged they're pre-profitability but emphasized their 4.2x net revenue retention and expanding contract sizes. [Tara Kapoor](people/tara-kapoor-111) pushed on the enterprise sales cycle question — Beta admitted deals take 3-4 months on average but noted they're experimenting with a self-serve tier for smaller teams.\n\nOverall sentiment from attendees was cautiously optimistic. The cybersecurity market is crowded but Beta's focus on the mid-market segment and their technical differentaition around behavioral analysis sets them apart. Carol mentioned she'd be scheduling a follow-up call with the founders next week to discuss potential term sheet parameters. The presentation ran about 12 minutes with 8 minutes of Q&A, slightly over the allocated time but nobody seemed to mind given the engaged discussion.",
-  "timeline": "- **2023-09-15** | Beta accepted into W25 batch, begins intensive program work\n- **2024-01-08** | [Carol Jackson](people/carol-jackson-81) first introduced to Beta team during office hours\n- **2024-02-02** | Beta crosses $800K ARR milestone ahead of demo day\n- **2024-02-14** | Practice pitch session with batch partners, feedback incorporated\n- **2024-02-16** | Demo Day W25 presentation to investor panel including [Victor Taylor](people/victor-taylor-1)\n- **2024-02-23** | Follow-up meeting scheduled between [Carol Jackson](people/carol-jackson-81) and Beta founders\n- **2024-03-11** | [Tara Kapoor](people/tara-kapoor-111) shares Beta deck with her fund partners\n- **2024-04-02** | Beta closes Series A discussions begin in earnest",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/demo-day-2024-02-16-batch-1",
-    "name": "Demo Day W25",
-    "meeting_type": "demo_day",
-    "date": "2024-02-16",
-    "attendees": [
-      "people/carol-jackson-81",
-      "people/victor-taylor-1",
-      "people/uma-brown-6",
-      "people/henry-johnson-12",
-      "people/tara-kapoor-111"
-    ],
-    "topic_company": "companies/beta-1"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__demo-day-2024-03-17-batch-2.json b/eval/data/world-v1/meetings__demo-day-2024-03-17-batch-2.json
deleted file mode 100644
index dcea93ffa..000000000
--- a/eval/data/world-v1/meetings__demo-day-2024-03-17-batch-2.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "meetings/demo-day-2024-03-17-batch-2",
-  "type": "meeting",
-  "title": "Demo Day W26 – Gamma Presentation",
-  "compiled_truth": "Demo Day W26 took place on March 17th, 2024, featuring [Gamma](companies/gamma-2) as the presenting company. The fintech startup showcased their latest product iteration to a small but engaged audience of investors and advisors. [Iris Lee](people/iris-lee-82) led the presentation alongside [Mark Jones](people/mark-jones-2), walking through Gamma's traction numbers and roadmap for the remainder of 2024.\n\nThe session kicked off with a 15-minute product demo showing Gamma's new merchant reconciliation tools. Mark handled most of the technical walkthrough while Iris fielded questions from the audience. [Ulrich Johnson](people/ulrich-johnson-7) asked pointed questions about unit economics, particularly around customer acquisiton costs in the SMB segment. The team's answers were solid if somewhat rehearsed—they'd clearly anticipated pushback on their burn rate.\n\n[Mia Lee](people/mia-lee-13) attended primarily as an observer but jumped in during the Q&A to probe Gamma's competitive positioning against legacy players like Bill.com and newer entrants. [Tina Jones](people/tina-jones-112) provided feedback on the pitch deck structure, noting that the team should lead with the problem statement more aggressivley before diving into solution details.\n\nKey metrics presented: $340K ARR (up from $180K at previous demo day), 47 active merchant accounts, and 94% monthly retention. The team announced they're targeting a $3M seed extension to close by end of Q2. Ulrich expressed preliminary interest in participating, contingent on seeing April numbers.\n\nDiscussion got heated briefly when Mark mentioned plans to expand into invoice factoring. Several attendees questioned whether this was too much scope creep for an early-stage company. Iris defended the decision, arguing that their existing merchant relationships gave them unique distribution for working capital products. The room remained skeptical but agreed to revisit after the team shares more detailed financials.\n\nOverall sentiment was cautiously optimistic. Gamma has made real progress since W25, though the path to profitability remains unclear.",
-  "timeline": "- **2023-09-14** | Gamma first introduced to the cohort during intake session\n- **2023-11-20** | [Mark Jones](people/mark-jones-2) joins Gamma as technical co-founder\n- **2024-01-08** | W25 Demo Day presentation; $180K ARR milestone announced\n- **2024-02-22** | [Iris Lee](people/iris-lee-82) presents at fintech founder meetup, generates investor interest\n- **2024-03-10** | Pre-demo day prep call with [Tina Jones](people/tina-jones-112) for pitch coaching\n- **2024-03-17** | W26 Demo Day; Gamma presents to audience of 5\n- **2024-03-19** | Follow-up meeting scheduled with [Ulrich Johnson](people/ulrich-johnson-7) to discuss seed extension\n- **2024-04-02** | Target date for Gamma to share updated financials with interested investors",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/demo-day-2024-03-17-batch-2",
-    "name": "Demo Day W26",
-    "meeting_type": "demo_day",
-    "date": "2024-03-17",
-    "attendees": [
-      "people/iris-lee-82",
-      "people/mark-jones-2",
-      "people/ulrich-johnson-7",
-      "people/mia-lee-13",
-      "people/tina-jones-112"
-    ],
-    "topic_company": "companies/gamma-2"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__demo-day-2024-04-18-batch-3.json b/eval/data/world-v1/meetings__demo-day-2024-04-18-batch-3.json
deleted file mode 100644
index febbdc61a..000000000
--- a/eval/data/world-v1/meetings__demo-day-2024-04-18-batch-3.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "meetings/demo-day-2024-04-18-batch-3",
-  "type": "meeting",
-  "title": "Demo Day W27 - Delta Biotech Presentation",
-  "compiled_truth": "Demo Day W27 took place on April 18, 2024, featuring [Delta](companies/delta-3) as the presenting company. The biotech startup showcased their synthetic biology platform to an audience of partners and potential investors. [David Zhang](people/david-zhang-83) led the presentation, walking through Delta's core technology for engineering microbial strains that produce rare pharmaceutical precursors at scale.\n\nThe session drew strong attendance from the investing team. [Victor Wilson](people/victor-wilson-3) asked pointed questions about Delta's regulatory pathway, specifically around FDA approval timelines for their lead compound. David handled the regulatory questions well, noting they'd engaged consultants from former FDA reviewers. [Yara Johnson](people/yara-johnson-8) focused her questions on unit economics and whether the fermentation process could achieve cost parity with traditional chemical synthesis within 18 months.\n\n[Vera Chen](people/vera-chen-14) provided useful context from her experience evaluating other synbio deals, noting that Delta's approach to strain optimization seemed more capital-efficient than competitors she'd seen. She flagged some concerns about IP protection given how much of thier methodology relies on publicly available CRISPR techniques. [Adam Lopez](people/adam-lopez-113) rounded out the Q&A by probing the team composition—Delta currently has 8 FTEs and plans to double headcount if they close their Series A.\n\nThe demo itself was impressive. Zhang showed real-time fermentation data from their pilot facility, demonstrating yield improvements of 340% over the past quarter. The team has clearly made technical progress since their seed round. Discussion turned to go-to-market strategy, with Delta targeting contract manufacturing deals with mid-size pharma companies who struggle to source certain precursors reliably.\n\nPost-presentation sentiment was generally positive. Victor noted this was one of the stronger biotech pitches from the batch. Follow-up diligence will focus on validating the IP moat and speaking with potential pharma customers. Next steps include a site visit to Delta's lab facility, tentatively scheduled for early May.",
-  "timeline": "- **2023-09-15** | [Delta](companies/delta-3) accepted into W27 batch after competitive application process\n- **2024-01-22** | Initial partner meeting with [David Zhang](people/david-zhang-83) to review Delta's technical roadmap\n- **2024-03-08** | Delta completes pilot fermentation run with breakthrough yield results\n- **2024-04-10** | Pre-demo day prep session held with [Vera Chen](people/vera-chen-14) providing feedback on pitch deck\n- **2024-04-18** | Demo Day W27 presentation delivered to full partner group\n- **2024-04-25** | [Victor Wilson](people/victor-wilson-3) schedules follow-up call with Delta to discuss term sheet parameters\n- **2024-05-03** | Site visit to Delta lab facility in South San Francisco\n- **2024-05-19** | Delta closes $12M Series A with participation from fund",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/demo-day-2024-04-18-batch-3",
-    "name": "Demo Day W27",
-    "meeting_type": "demo_day",
-    "date": "2024-04-18",
-    "attendees": [
-      "people/david-zhang-83",
-      "people/victor-wilson-3",
-      "people/yara-johnson-8",
-      "people/vera-chen-14",
-      "people/adam-lopez-113"
-    ],
-    "topic_company": "companies/delta-3"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__demo-day-2024-05-19-batch-4.json b/eval/data/world-v1/meetings__demo-day-2024-05-19-batch-4.json
deleted file mode 100644
index 651e1a6b1..000000000
--- a/eval/data/world-v1/meetings__demo-day-2024-05-19-batch-4.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "meetings/demo-day-2024-05-19-batch-4",
-  "type": "meeting",
-  "title": "Demo Day W28 – Epsilon Pitch",
-  "compiled_truth": "Demo Day W28 featured [Epsilon](companies/epsilon-4) presenting their cybersecurity platform to a small but engaged audience of investors and advisors. The session took place on May 19th, 2024, with [Sarah Lopez](people/sarah-lopez-84) leading the pitch on behalf of the Epsilon team. She walked through their core product—an AI-driven threat detection system designed for mid-market enterprises that lack dedicated security operations centers.\n\nThe room included [Paul Rodriguez](people/paul-rodriguez-4), who asked pointed questions about Epsilon's go-to-market strategy and whether they'd considered channel partnerships with MSPs. [Rachel Garcia](people/rachel-garcia-9) was particularly interested in the technical architecture, probing on false positive rates and how the model handles zero-day threats. Her background in enterprise software made her skeptical of some of the performance claims, though she acknowledged the demo was impressive.\n\n[Noah Kapoor](people/noah-kapoor-15) attended in an advisory capacity and offered feedback on the pitch deck structure—he felt the team buried the lede on their traction numbers, which were actually quite strong for a seed-stage company. [Julia Johnson](people/julia-johnson-114) rounded out the attendees, taking notes and asking about Epsilon's hiring plans for the next two quarters.\n\nKey discussion points included pricing strategy (Epsilon currently charges per endpoint, but there was debate about whether a flat-rate model would reduce friction), competitive positioning against CrowdStrike and SentinelOne, and the regulatory tailwinds from new SEC disclosure requirements. Sarah handled the Q&A well, though she stumbled a bit when pressed on unit economics—something the team acknowledged they need to tighten up before their Series A conversations.\n\nOverall sentiment was cautiously optimistic. The product clearly has legs, and the founding team's pedigree (two ex-Palo Alto Networks engineers) gives them credibilty in a crowded space. Follow-up meetings were scheduled with both Paul and Rachel for deeper dives.",
-  "timeline": "- **2024-03-01** | [Epsilon](companies/epsilon-4) applies to W28 batch, submits initial application video\n- **2024-04-15** | Epsilon accepted into batch, begins intensive prep for demo day\n- **2024-05-10** | [Sarah Lopez](people/sarah-lopez-84) runs pitch practice with batch mentors\n- **2024-05-19** | Demo Day W28 presentation; strong Q&A with [Rachel Garcia](people/rachel-garcia-9) and [Paul Rodriguez](people/paul-rodriguez-4)\n- **2024-05-22** | Follow-up meeting scheduled between Epsilon and Paul Rodriguez\n- **2024-06-03** | [Noah Kapoor](people/noah-kapoor-15) sends deck feedback and intro to potential angel investors\n- **2024-06-18** | Epsilon closes bridge round; terms influenced by demo day conversations",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/demo-day-2024-05-19-batch-4",
-    "name": "Demo Day W28",
-    "meeting_type": "demo_day",
-    "date": "2024-05-19",
-    "attendees": [
-      "people/sarah-lopez-84",
-      "people/paul-rodriguez-4",
-      "people/rachel-garcia-9",
-      "people/noah-kapoor-15",
-      "people/julia-johnson-114"
-    ],
-    "topic_company": "companies/epsilon-4"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__demo-day-2024-06-20-batch-5.json b/eval/data/world-v1/meetings__demo-day-2024-06-20-batch-5.json
deleted file mode 100644
index 31aefdc1b..000000000
--- a/eval/data/world-v1/meetings__demo-day-2024-06-20-batch-5.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "meetings/demo-day-2024-06-20-batch-5",
-  "type": "meeting",
-  "title": "Demo Day W29 – Nimbus Pitch",
-  "compiled_truth": "Demo Day W29 featured [Nimbus](companies/nimbus-5) presenting their climate tech platform to a mixed audience of partners and potential co-investors. The session ran about 45 minutes including Q&A, held in the main conference room with video conferencing for remote attendees.\n\n[Priya Taylor](people/priya-taylor-85) led the presentation, walking through Nimbus's core product—an AI-powered carbon accounting system targeting mid-market manufacturing companies. She emphasized their recent traction with three pilot customers and a 40% month-over-month growth in data processed. The pitch was polished, though some slides felt a bit dense with technical details that might've lost generalist investors.\n\n[David Wang](people/david-wang-10) asked pointed questions about unit economics and the path to profitability. Priya handled these well, noting they expect to hit breakeven within 18 months assuming current growth holds. [Mia Anderson](people/mia-anderson-5) probed on competitive landscape—specifically how Nimbus differentiates from Watershed and Persefoni. The answer centered on their focus on manufacturing verticals and proprietary sensor integrations, which seemed to satisfy most concerns.\n\n[Ulrich Wang](people/ulrich-wang-16) provided feedback on go-to-market strategy, suggesting they consider channel partnerships with existing ERP vendors rather than pure direct sales. This sparked a longer discussion about integration complexity and sales cycle lengths. [Quinten Nakamura](people/quinten-nakamura-115) was quieter during the session but followed up afterward with questions about the technical architecture.\n\nOverall sentiment in the room was cautiously optimistic. The team acknowledged Nimbus still needs to prove enterprise sales repeatability, but the product-market fit signals are encouraging. Climate tech remains a crowded space but Nimbus's vertical focus could be a real differentiator if they execute well. Follow-up diligence scheduled for next week to review financials and customer references in more detial.",
-  "timeline": "- **2024-06-15** | Nimbus submitted final pitch deck for Demo Day review\n- **2024-06-18** | Pre-demo prep call with [Priya Taylor](people/priya-taylor-85) to refine talking points\n- **2024-06-20** | Demo Day W29 held; [Nimbus](companies/nimbus-5) presented to full partner group\n- **2024-06-20** | [David Wang](people/david-wang-10) flagged unit economics concerns during Q&A\n- **2024-06-21** | [Mia Anderson](people/mia-anderson-5) circulated internal memo summarizing demo impressions\n- **2024-06-24** | Follow-up diligence call scheduled with Nimbus finance team\n- **2024-06-27** | [Quinten Nakamura](people/quinten-nakamura-115) completed technical architecture review",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/demo-day-2024-06-20-batch-5",
-    "name": "Demo Day W29",
-    "meeting_type": "demo_day",
-    "date": "2024-06-20",
-    "attendees": [
-      "people/priya-taylor-85",
-      "people/mia-anderson-5",
-      "people/david-wang-10",
-      "people/ulrich-wang-16",
-      "people/quinten-nakamura-115"
-    ],
-    "topic_company": "companies/nimbus-5"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__demo-day-2024-07-21-batch-6.json b/eval/data/world-v1/meetings__demo-day-2024-07-21-batch-6.json
deleted file mode 100644
index 0eea631a9..000000000
--- a/eval/data/world-v1/meetings__demo-day-2024-07-21-batch-6.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "meetings/demo-day-2024-07-21-batch-6",
-  "type": "meeting",
-  "title": "Demo Day W30",
-  "compiled_truth": "Demo Day W30 took place on July 21, 2024, showcasing the sixth batch of startups from the accelerator program. The event brought together investors, mentors, and founders for a day of pitches and networking. [Vector](companies/vector-6) was the featured company for this session, presenting their health tech platform to a room of engaged stakeholders.\n\nThe attending group included [Julia Davis](people/julia-davis-86), who led the proceedings and provided opening remarks about the batch's overall progress. [Uma Brown](people/uma-brown-6) attended in her capacity as a potential lead investor, having expressed prior intrest in health tech opporunities. Mark Thomas was there representing his firm's scout program, while [Quinten Wang](people/quinten-wang-17) came specifically to evaluate Vector's technical architecture.\n\nVictor Jackson rounded out the attendee list. He'd been following Vector since their seed announcement and wanted to see how the product had evolved over the past quarter.\n\nThe Vector team delivered a 12-minute pitch followed by an extended Q&A. Their presentation focused on recent traction metrics, showing 340% month-over-month growth in provider sign-ups. The founders addressed questions about their go-to-market strategy and regulatory pathway, which seemed to satisfy most concerns from the investor side.\n\nKey discussion points centered on Vector's competitive moat. Uma Brown pushed hard on unit economics, asking pointed questions about customer acquisition costs in the healthcare vertical. Julia Davis facilitated a productive back-and-forth about potential strategic partnerships with existing EHR providers.\n\nThe mood was generally optimistic. Several attendees noted this was one of the stronger demo day performances from the batch. Post-presentation conversations continued for nearly two hours, with multiple attendees exchanging contact information with the Vector founding team. Follow-up meetings were scheduled for the subsequent week.",
-  "timeline": "- **2024-06-15** | Vector confirmed as featured company for W30 Demo Day\n- **2024-07-01** | [Julia Davis](people/julia-davis-86) sent calendar invites to confirmed attendees\n- **2024-07-18** | Pre-demo day dinner held for mentors and batch founders\n- **2024-07-21** | Demo Day W30 held; [Vector](companies/vector-6) pitched to investor panel\n- **2024-07-21** | [Uma Brown](people/uma-brown-6) requested follow-up meeting with Vector team\n- **2024-07-28** | Post-demo feedback session conducted with batch companies\n- **2024-08-03** | [Quinten Wang](people/quinten-wang-17) completed technical due diligence call with Vector CTO",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/demo-day-2024-07-21-batch-6",
-    "name": "Demo Day W30",
-    "meeting_type": "demo_day",
-    "date": "2024-07-21",
-    "attendees": [
-      "people/julia-davis-86",
-      "people/uma-brown-6",
-      "people/mark-thomas-11",
-      "people/quinten-wang-17",
-      "people/victor-jackson-116"
-    ],
-    "topic_company": "companies/vector-6"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__demo-day-2024-08-22-batch-7.json b/eval/data/world-v1/meetings__demo-day-2024-08-22-batch-7.json
deleted file mode 100644
index c96a82a22..000000000
--- a/eval/data/world-v1/meetings__demo-day-2024-08-22-batch-7.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "meetings/demo-day-2024-08-22-batch-7",
-  "type": "meeting",
-  "title": "Demo Day W31 – Quantum",
-  "compiled_truth": "Demo Day W31 featured [Quantum](companies/quantum-7) presenting their fintech platform to a small but engaged group of investors and advisors. The session took place on August 22nd, 2024, with [Helen Martinez](people/helen-martinez-87) leading the presentation on behalf of the company. She walked through Quantum's core value proposition—a real-time payments reconciliation engine designed for mid-market e-commerce companies struggling with fragmented payment processor data.\n\nThe attendee list was tight: [Ulrich Johnson](people/ulrich-johnson-7) came in with pointed questions about unit economics, pressing Helen on whether the 2.3% take rate was sustainable given competitor pricing. [Henry Johnson](people/henry-johnson-12) seemed more interested in the technical architecture, asking about latency benchmarks and whether the team had stress-tested the system during peak shopping periods. His questions revealed some skepticism about the infrastructure choices but he acknowledged the team's execution speed.\n\n[Nina Rodriguez](people/nina-rodriguez-18) focused on go-to-market, noting that Quantum's current customer concentration (three clients representing 70% of revenue) posed a risk. She suggested exploring channel partnerships with accounting software providers as a distribution lever. [Tina Lopez](people/tina-lopez-117) mostly observed but jumped in toward the end with questions about the competitive moat—specifically, what prevents Stripe or Plaid from building this natively.\n\nHelen handled the tough questions well, though she stumbled a bit when pressed on churn numbers from Q2. The demo itself was polished, showing a merchant dashboard that aggregates PayPal, Stripe, and Square transactions into a unified reconcilliation view. Response from the room was cautiously positive. Ulrich mentioned he'd want to see another quarter of data before committing, while Nina expressed interest in a follow-up call to discuss the Series A timeline.\n\nKey takeaway from the session: Quantum has product-market fit signals but needs to diversify its customer base and tighten the narrative around defensibility. The team is targeting a $4M raise in Q4 2024.",
-  "timeline": "- **2024-06-15** | [Quantum](companies/quantum-7) selected for Batch 7 accelerator program\n- **2024-07-10** | [Helen Martinez](people/helen-martinez-87) completes pitch coaching sessions with program mentors\n- **2024-08-01** | Quantum closes pilot with fourth enterprise customer ahead of demo day\n- **2024-08-22** | Demo Day W31 presentation; [Ulrich Johnson](people/ulrich-johnson-7) and [Nina Rodriguez](people/nina-rodriguez-18) attend\n- **2024-08-25** | Follow-up meeting scheduled between Helen and Nina to discuss Series A\n- **2024-09-12** | [Tina Lopez](people/tina-lopez-117) sends intro to potential strategic partner\n- **2024-11-03** | Quantum begins Series A roadshow targeting $4M raise",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/demo-day-2024-08-22-batch-7",
-    "name": "Demo Day W31",
-    "meeting_type": "demo_day",
-    "date": "2024-08-22",
-    "attendees": [
-      "people/helen-martinez-87",
-      "people/ulrich-johnson-7",
-      "people/henry-johnson-12",
-      "people/nina-rodriguez-18",
-      "people/tina-lopez-117"
-    ],
-    "topic_company": "companies/quantum-7"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__demo-day-2024-09-23-batch-8.json b/eval/data/world-v1/meetings__demo-day-2024-09-23-batch-8.json
deleted file mode 100644
index 4e3175136..000000000
--- a/eval/data/world-v1/meetings__demo-day-2024-09-23-batch-8.json
+++ /dev/null
@@ -1,31 +0,0 @@
-{
-  "slug": "meetings/demo-day-2024-09-23-batch-8",
-  "type": "meeting",
-  "title": "Demo Day W32 – Pulse",
-  "compiled_truth": "Demo Day W32 took place on September 23, 2024, featuring [Pulse](companies/pulse-8) as the presenting company. The session was part of Batch 8's culminating presentations, where founders showcase their progress to partners and invited guests. Pulse, an edtech startup focused on real-time student engagement analytics, demonstrated significant traction since entering the program.\n\n[Fiona Moore](people/fiona-moore-88) led the presentation, walking through Pulse's core product: a classroom dashboard that helps teachers identify struggling students before they fall behind. She emphasized the company's recent pilot deployments across 12 school districts in the Pacific Northwest, noting a 34% improvement in early intervention rates. The metrics were compelling—retention among pilot schools sat at 89%, and several districts had already signed annual contracts.\n\n[Yara Johnson](people/yara-johnson-8) handled the Q&A portion alongside Fiona, fielding questions about data privacy compliance and the competitive landscape. One notable exchange involved [Xavier Nakamura](people/xavier-nakamura-118), who pushed back on the unit economics, questioning whether the $8/student/year pricing could sustain the level of customer sucess required for K-12 sales. Fiona acknowledged the concern but pointed to their recently hired sales team and partnerships with two state education departments.\n\n[Mia Lee](people/mia-lee-13) attended as an observer and later provided feedback on the pitch deck structure, suggesting they lead with the emotional hook—a teacher testimonial video—rather than market size slides. [Adam Lee](people/adam-lee-19) was also present, though he arrived late and missed the first ten minutes of the demo. He did contribute during the breakout discussion afterward, recommending Pulse consider enterprise licensing for charter school networks.\n\nOverall sentiment from attendees was positive. The team's execution speed impressed partners, particularly the fact that they'd closed $180K in ARR within eight weeks of launching. Some concerns remain around scaling their implementation support, but Demo Day W32 solidified Pulse as one of the standout companies from this cohort.",
-  "timeline": [
-    "- **2024-06-15** | [Pulse](companies/pulse-8) accepted into Batch 8 accelerator program",
-    "- **2024-07-22** | First pilot deployment begins in Seattle-area school district",
-    "- **2024-08-10** | [Fiona Moore](people/fiona-moore-88) presents mid-batch update to partners",
-    "- **2024-09-01** | Pulse crosses $100K ARR milestone",
-    "- **2024-09-18** | Demo Day rehearsal session with feedback from [Mia Lee](people/mia-lee-13)",
-    "- **2024-09-23** | Demo Day W32 presentation delivered to partners and guests",
-    "- **2024-09-25** | Follow-up meeting scheduled with [Xavier Nakamura](people/xavier-nakamura-118) to discuss potential investment",
-    "- **2024-10-02** | Post-demo investor interest summary circulated to Pulse team"
-  ],
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/demo-day-2024-09-23-batch-8",
-    "name": "Demo Day W32",
-    "meeting_type": "demo_day",
-    "date": "2024-09-23",
-    "attendees": [
-      "people/fiona-moore-88",
-      "people/yara-johnson-8",
-      "people/mia-lee-13",
-      "people/adam-lee-19",
-      "people/xavier-nakamura-118"
-    ],
-    "topic_company": "companies/pulse-8"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__demo-day-2024-10-24-batch-9.json b/eval/data/world-v1/meetings__demo-day-2024-10-24-batch-9.json
deleted file mode 100644
index fb8c9740b..000000000
--- a/eval/data/world-v1/meetings__demo-day-2024-10-24-batch-9.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "meetings/demo-day-2024-10-24-batch-9",
-  "type": "meeting",
-  "title": "Demo Day W33 – Helix Pitch",
-  "compiled_truth": "Demo Day W33 featured [Helix](companies/helix-9) presenting their AI infrastructure platform to a small but engaged group of partners. The session took place on October 24, 2024, with [Jack Davis](people/jack-davis-89) leading the presentation alongside his technical co-founder. The room included [Rachel Garcia](people/rachel-garcia-9), [Vera Chen](people/vera-chen-14), [Vera Singh](people/vera-singh-20), and [Quinn Park](people/quinn-park-119).\n\nHelix's pitch focused on their managed inference layer—essentially a routing system that sits between applications and multiple LLM providers. Jack walked through their core value prop: developers integrate once, and Helix handles failover, cost optimization, and latency routing across OpenAI, Anthropic, and open-source models. The technical demo was solid, showing real-time switching when one provider hit rate limits.\n\n[Quinn Park](people/quinn-park-119) pushed back hard on moat. \"What stops the big providers from just building this themselves?\" Jack's response leaned on their observability layer—they're collecting data on actual production usage patterns that lets them predict optimal routing before failures happen. Vera Singh seemed convinced by this, noting that similar infrastructure plays in adjacent spaces have proven sticky once they hit scale.\n\nDiscussion turned to go-to-market. Helix is currently free up to 1M tokens/month, then charges a small markup on passthrough costs. They've got 340 developers using the platform, 12 paying customers. Revenue is thin—around $8k MRR—but growing 40% month over month. Rachel asked about enterprise interest; Jack mentioned two Fortune 500 pilots but couldn't share names yet.\n\nThe team seemed strong. Jack handles product and sales, his co-founder (former Google infra engineer) owns the technical side. They're looking to raise a seed round, targeting $3M at $15M post. General consensus in the room was positive but cautious—the space is moving fast and consolidation risk is real. [Vera Chen](people/vera-chen-14) offered to make intros to a couple portfolio companies that might be good design partners.",
-  "timeline": "- **2024-06-15** | Helix founded by [Jack Davis](people/jack-davis-89) and CTO, initial prototype built\n- **2024-08-01** | Launched free tier, first 50 developers sign up within two weeks\n- **2024-09-10** | Hit 200 active developers on platform, first paying customer converts\n- **2024-10-03** | Reached $8k MRR milestone, 12 paying accounts\n- **2024-10-24** | Demo Day W33 presentation to [Quinn Park](people/quinn-park-119) and partner group\n- **2024-11-02** | Follow-up meeting scheduled with [Vera Chen](people/vera-chen-14) for portfolio intros\n- **2024-12-15** | Target close date for seed round discussions\n- **2025-02-01** | Planned launch of enterprise tier with SLA guarantees",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/demo-day-2024-10-24-batch-9",
-    "name": "Demo Day W33",
-    "meeting_type": "demo_day",
-    "date": "2024-10-24",
-    "attendees": [
-      "people/jack-davis-89",
-      "people/rachel-garcia-9",
-      "people/vera-chen-14",
-      "people/vera-singh-20",
-      "people/quinn-park-119"
-    ],
-    "topic_company": "companies/helix-9"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__demo-day-2024-11-15-batch-10.json b/eval/data/world-v1/meetings__demo-day-2024-11-15-batch-10.json
deleted file mode 100644
index 197b4981a..000000000
--- a/eval/data/world-v1/meetings__demo-day-2024-11-15-batch-10.json
+++ /dev/null
@@ -1,31 +0,0 @@
-{
-  "slug": "meetings/demo-day-2024-11-15-batch-10",
-  "type": "meeting",
-  "title": "Demo Day W34 – Beacon Pitch",
-  "compiled_truth": "Demo Day W34 took place on November 15, 2024, featuring [Beacon](companies/beacon-10) as the presenting company. The cybersecurity startup pitched their autonomous threat detection platform to a panel of investors and partners. [Rosa Jackson](people/rosa-jackson-90) led the presentation, walking through Beacon's core technology and go-to-market strategy with her usual directness.\n\nThe room included [David Wang](people/david-wang-10), [Noah Kapoor](people/noah-kapoor-15), [Eric Lee](people/eric-lee-21), and [Ulrich Kim](people/ulrich-kim-120). David asked pointed questions about Beacon's enterprise sales cycle, noting that their 90-day average seemed long for the current market. Rosa pushed back, explaining that security procurement inherently involves longer cycles due to compliance requirements. Noah seemed particularly interested in the AI componenets of the threat detection system, asking several follow-ups about false positive rates.\n\nEric raised concerns about competitive positioning against CrowdStrike and Palo Alto Networks. The Beacon team acknowledged the challenge but emphasized their focus on mid-market companies that find enterprise solutions too expensive and complex. Ulrich asked about burn rate and runway—Beacon disclosed they have approximately 14 months remaining at current spend.\n\nKey discussion points centered on: customer acquisition costs (currently ~$18k per enterprise customer), the technical moat around their behavioral analysis engine, and expansion revenue from existing accounts. Rosa mentioned they're seeing 130% net dollar retention, which caught David's attention. He noted that's comparable to best-in-class SaaS metrics.\n\nThe demo itself went smoothly, showing real-time threat detection across a simulated network. One minor hiccup when the dashboard lagged during the live portion, but Rosa handled it gracefully. Overall reception was positive though not overwhelmingly enthusiastic. Several attendees mentioned wanting to see more traction before commiting to follow-on conversations. Noah suggested reconnecting in Q1 2025 after Beacon closes a few more enterprise deals.",
-  "timeline": [
-    "- **2024-10-28** | Beacon confirmed as presenting company for Demo Day W34",
-    "- **2024-11-01** | [Rosa Jackson](people/rosa-jackson-90) submitted pitch deck for review",
-    "- **2024-11-08** | Pre-demo prep call with [Noah Kapoor](people/noah-kapoor-15) to refine messaging",
-    "- **2024-11-12** | Final attendee list confirmed; [Ulrich Kim](people/ulrich-kim-120) added last minute",
-    "- **2024-11-15** | Demo Day W34 held; Beacon presented 25-minute pitch plus Q&A",
-    "- **2024-11-18** | Follow-up notes distributed to all attendees",
-    "- **2024-11-22** | [David Wang](people/david-wang-10) requested additional financial materials from Beacon",
-    "- **2024-12-03** | Scheduled check-in call between Rosa and Eric to discuss potential partnership angles"
-  ],
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/demo-day-2024-11-15-batch-10",
-    "name": "Demo Day W34",
-    "meeting_type": "demo_day",
-    "date": "2024-11-15",
-    "attendees": [
-      "people/rosa-jackson-90",
-      "people/david-wang-10",
-      "people/noah-kapoor-15",
-      "people/eric-lee-21",
-      "people/ulrich-kim-120"
-    ],
-    "topic_company": "companies/beacon-10"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__demo-day-2024-12-16-batch-11.json b/eval/data/world-v1/meetings__demo-day-2024-12-16-batch-11.json
deleted file mode 100644
index fad993b9e..000000000
--- a/eval/data/world-v1/meetings__demo-day-2024-12-16-batch-11.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "meetings/demo-day-2024-12-16-batch-11",
-  "type": "meeting",
-  "title": "Demo Day W35 – Compass",
-  "compiled_truth": "Demo Day W35 took place on December 16, 2024, featuring [Compass](companies/compass-11) as the presenting company. The session brought together a focused group of attendees including [Chris Jackson](people/chris-jackson-91), [Mark Thomas](people/mark-thomas-11), [Ulrich Wang](people/ulrich-wang-16), [Quinten Rodriguez](people/quinten-rodriguez-22), and [Rachel Davis](people/rachel-davis-121). This was batch 11's final demo before the holiday break.\n\nCompass presented their latest iteration of the cross-chain settlement protocol. The team walked through a live demonstration of atomic swaps between Ethereum and Solana, completing a $50k equivalent transaction in under 4 seconds. [Mark Thomas](people/mark-thomas-11) asked pointed questions about MEV protection, which led to a fifteen-minute deep dive into their sequencer architecture. The Compass team handled it well, showing they'd thought through the adversarial cases.\n\nChris Jackson raised concerns about regulatory exposure given recent SEC actions in the crypto space. The discussion got heated — [Quinten Rodriguez](people/quinten-rodriguez-22) pushed back, arguing that infrastructure plays are fundementally different from token issuance. Rachel Davis mediated, noting that the legal memo from batch 9 might be worth revisiting. Action item assigned to follow up with outside counsel.\n\nTraction metrics were encouraging. Compass showed $2.3M in monthly volume, up from $800k at the previous demo day. User retention sitting at 34% weekly, which is strong for DeFi tooling. [Ulrich Wang](people/ulrich-wang-16) commented that the unit economics still look upside down but acknowledged the land-grab dynamics of the space.\n\nThe team requested introductions to three specific LPs for their upcoming seed extension. Rachel committed to making two of those intros by end of week. Overall sentiment was cautiosly optimistic — the product is clearly working, the question is whether the market timing is right given macro headwinds in crypto.",
-  "timeline": "- **2024-10-14** | Compass first pitched at Demo Day W27, received feedback to focus on single-chain before expanding\n- **2024-11-04** | [Mark Thomas](people/mark-thomas-11) conducted technical due diligence call with Compass engineering team\n- **2024-11-18** | Compass crossed $1M monthly volume milestone\n- **2024-12-02** | Pre-demo check-in with [Rachel Davis](people/rachel-davis-121) to review pitch deck updates\n- **2024-12-16** | Demo Day W35 presentation, batch 11 final demo before holidays\n- **2024-12-19** | Follow-up intros scheduled to be sent by Rachel\n- **2025-01-06** | Next demo day scheduled for batch 12 kickoff",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/demo-day-2024-12-16-batch-11",
-    "name": "Demo Day W35",
-    "meeting_type": "demo_day",
-    "date": "2024-12-16",
-    "attendees": [
-      "people/chris-jackson-91",
-      "people/mark-thomas-11",
-      "people/ulrich-wang-16",
-      "people/quinten-rodriguez-22",
-      "people/rachel-davis-121"
-    ],
-    "topic_company": "companies/compass-11"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__demo-day-2025-01-17-batch-12.json b/eval/data/world-v1/meetings__demo-day-2025-01-17-batch-12.json
deleted file mode 100644
index 301488bca..000000000
--- a/eval/data/world-v1/meetings__demo-day-2025-01-17-batch-12.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "meetings/demo-day-2025-01-17-batch-12",
-  "type": "meeting",
-  "title": "Demo Day W36 — Lumen (Batch 12)",
-  "compiled_truth": "Demo Day W36 featured [Lumen](companies/lumen-12) presenting their biotech platform to the batch 12 cohort. The session was held on January 17th, 2025, with a solid turnout from partners and fellow founders. [Sarah Williams](people/sarah-williams-92) led the presentation, walking through Lumen's core technology—a novel approach to protein folding prediction that they claim outperforms existing methods on certain edge cases.\n\nThe demo itself ran about 18 minutes. Sarah showed live inference on a handful of protein sequences, highlighting where their model diverges from AlphaFold predictions and why they beleive their approach captures dynamics better. [Henry Johnson](people/henry-johnson-12) jumped in during Q&A to push back on the training data methodology, asking whether they'd validated against wet lab results or just computational benchmarks. Good question—Lumen's team acknowledged they're still working on securing pharma partnerships for real-world validation.\n\n[Quinten Wang](people/quinten-wang-17) asked about go-to-market, specifically whether they're targeting drug discovery companies directly or planning to license the tech. Sarah mentioned they're in early conversations with two mid-size biotechs but couldn't share names yet. [Paul Anderson](people/paul-anderson-23) seemed particularly interested in the infrastructure side, asking about compute costs and whether they'd considered spinning off the training pipeline as a seperate product.\n\n[Grace Miller](people/grace-miller-122) provided closing feedback, noting that the technical depth was impressive but the pitch could use more clarity on the business model. She recommended they tighten the narrative before investor meetings next month. Overall, the room was engaged—lots of follow-up questions after the formal session ended. Lumen is clearly one of the more technically ambitious companies in this batch, though execution risk remains high given the competitive landscape in computational biology.",
-  "timeline": "- **2024-09-15** | [Lumen](companies/lumen-12) accepted into Batch 12 accelerator program\n- **2024-10-22** | [Sarah Williams](people/sarah-williams-92) presents initial protein folding research at internal workshop\n- **2024-11-08** | Lumen completes first benchmark comparison against AlphaFold\n- **2024-12-03** | Team expands with hire of ML infrastructure engineer\n- **2025-01-10** | Demo Day prep session with [Grace Miller](people/grace-miller-122)\n- **2025-01-17** | Demo Day W36 presentation to Batch 12 cohort\n- **2025-01-24** | Follow-up meeting scheduled with [Henry Johnson](people/henry-johnson-12) on validation methodology\n- **2025-02-12** | Investor roadshow planned to begin",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/demo-day-2025-01-17-batch-12",
-    "name": "Demo Day W36",
-    "meeting_type": "demo_day",
-    "date": "2025-01-17",
-    "attendees": [
-      "people/sarah-williams-92",
-      "people/henry-johnson-12",
-      "people/quinten-wang-17",
-      "people/paul-anderson-23",
-      "people/grace-miller-122"
-    ],
-    "topic_company": "companies/lumen-12"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__demo-day-2025-02-18-batch-13.json b/eval/data/world-v1/meetings__demo-day-2025-02-18-batch-13.json
deleted file mode 100644
index e86382b93..000000000
--- a/eval/data/world-v1/meetings__demo-day-2025-02-18-batch-13.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "meetings/demo-day-2025-02-18-batch-13",
-  "type": "meeting",
-  "title": "Demo Day W37 – Cipher Batch 13 Presentation",
-  "compiled_truth": "Demo Day W37 took place on February 18, 2025, featuring [Cipher](companies/cipher-13) as the spotlight company for Batch 13's weekly showcase. The session drew a solid turnout with five attendees gathered to evaluate the fintech startup's progress and provide feedback on their recent development sprint.\n\n[Eric Martinez](people/eric-martinez-93) led the proceedings, kicking things off with a quick recap of Cipher's positioning in the embedded payments space. The team has been iterating rapidly on their merchant onboarding flow, which was the primary focus of this demo. Mia Lee from the Cipher team walked through the new streamlined KYC process—what used to take 3-4 days is now down to under 6 hours for most small businesses. Impressive stuff, though [Nina Rodriguez](people/nina-rodriguez-18) raised some pointed questions about edge cases involving international merchants.\n\nQuinten Lee provided technical feedback on the API architecture, noting that the webhook reliability had improved substiantally since the last demo. He suggested they consider adding idempotency keys to prevent duplicate transaction issues, which the Cipher team noted for their backlog. The discussion got a bit heated when Julia brought up competitve positioning—specifically how Cipher plans to differentiate from Stripe's recent moves into the same SMB segment.\n\n[Julia Thomas](people/julia-thomas-123) pushed hard on unit economics, asking about customer acquisition costs and lifetime value projections. The Cipher founders acknowledged they're still in the \"land grab\" phase but shared early cohort data showing promising retention numbers. Monthly churn sitting around 2.3% which is better than industry average for their segment.\n\nThe meeting wrapped with action items: Cipher to prepare a deeper dive on their fraud detection capabilities for next week, and Eric volunteered to make intros to two potential enterprise pilot customers. Overall sentiment was positive—the team's execution velocity has clearly picked up since joining the batch. Nina mentioned she'd be following up offline about potential synergies with another portfolio company working on compliance tooling.",
-  "timeline": "- **2025-01-14** | Cipher accepted into Batch 13 accelerator program\n- **2025-01-28** | Initial demo day presentation by [Mia Lee](people/mia-lee-13) on MVP product\n- **2025-02-04** | [Eric Martinez](people/eric-martinez-93) assigned as primary mentor for Cipher\n- **2025-02-11** | Technical review session with [Quinten Lee](people/quinten-lee-24) on API architecture\n- **2025-02-18** | Demo Day W37 presentation featuring streamlined KYC demo\n- **2025-02-19** | Follow-up meeting scheduled with [Nina Rodriguez](people/nina-rodriguez-18) on compliance partnerships\n- **2025-02-25** | Planned deep-dive on fraud detection capabilities",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/demo-day-2025-02-18-batch-13",
-    "name": "Demo Day W37",
-    "meeting_type": "demo_day",
-    "date": "2025-02-18",
-    "attendees": [
-      "people/eric-martinez-93",
-      "people/mia-lee-13",
-      "people/nina-rodriguez-18",
-      "people/quinten-lee-24",
-      "people/julia-thomas-123"
-    ],
-    "topic_company": "companies/cipher-13"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__demo-day-2025-03-19-batch-14.json b/eval/data/world-v1/meetings__demo-day-2025-03-19-batch-14.json
deleted file mode 100644
index 68a8b3772..000000000
--- a/eval/data/world-v1/meetings__demo-day-2025-03-19-batch-14.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "meetings/demo-day-2025-03-19-batch-14",
-  "type": "meeting",
-  "title": "Demo Day W38 – Mosaic Pitch",
-  "compiled_truth": "Demo Day W38 brought together a small but focused group to watch [Mosaic](companies/mosaic-14) present their consumer social platform. The session was held on March 19, 2025, with [Rosa Nakamura](people/rosa-nakamura-94) leading the evaluation alongside [Vera Chen](people/vera-chen-14), [Adam Lee](people/adam-lee-19), [Vera Wilson](people/vera-wilson-25), and [Chris Rodriguez](people/chris-rodriguez-124).\n\nMosaic pitched their vision for a new kind of social network built around collaborative mood boards and shared aesthetic curation. The core thesis is that Gen Z users are fatigued by performative posting and want spaces that feel more like co-creation than broadcasting. Their demo showed real-time collaborative boards where friends can drop images, songs, and vibes into shared canvases. Chris Rodriguez asked pointed questions about retention metrics—specifically whether users return after the initial novelty wears off. The Mosaic team shared early cohort data showing 34% D30 retention, which Rosa noted was promising but not yet at the threshold they typically look for.\n\n[Vera Chen](people/vera-chen-14) raised concerns about monetization pathways. The team mentioned potential brand partnerships and a premium tier for advanced curation tools, though admitted the business model is still evolving. Adam Lee seemed more optimistic, drawing comparisons to Pinterest's early days and noting that consumer social often takes longer to find its footing.\n\nDiscussion turned to competitive landscape. Vera Wilson pointed out that several startups are chasing the same \"anti-Instagram\" positioning, and differentiation will be key. The Mosaic founders emphasized their unique angle on synchronous collaboration—most competitors are async-first.\n\nOverall sentiment was cautiously positive. Rosa summarized the group's take: strong product intuition, compelling early traction, but needs another quarter of data before a conviction investment. Follow-up meeting tentatively scheduled for late April to review updated metrics.",
-  "timeline": "- **2024-09-12** | First intorduction to [Mosaic](companies/mosaic-14) via warm intro from a portfolio founder\n- **2024-11-03** | [Rosa Nakamura](people/rosa-nakamura-94) attends Mosaic's private beta launch event in Brooklyn\n- **2025-01-15** | Initial screening call with Mosaic founding team\n- **2025-02-20** | [Vera Chen](people/vera-chen-14) reviews Mosaic's product deck and financial projections\n- **2025-03-05** | Mosaic confirmed for Demo Day W38 presentation slot\n- **2025-03-19** | Demo Day W38 held; Mosaic presents to evaluation committee\n- **2025-03-21** | Internal debrief notes circulated among attendees\n- **2025-04-28** | Tentative follow-up scheduled to review Q1 retention data",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/demo-day-2025-03-19-batch-14",
-    "name": "Demo Day W38",
-    "meeting_type": "demo_day",
-    "date": "2025-03-19",
-    "attendees": [
-      "people/rosa-nakamura-94",
-      "people/vera-chen-14",
-      "people/adam-lee-19",
-      "people/vera-wilson-25",
-      "people/chris-rodriguez-124"
-    ],
-    "topic_company": "companies/mosaic-14"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-0-2025-01-01.json b/eval/data/world-v1/meetings__oneonone-0-2025-01-01.json
deleted file mode 100644
index 4b1ccc02f..000000000
--- a/eval/data/world-v1/meetings__oneonone-0-2025-01-01.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-0-2025-01-01",
-  "type": "meeting",
-  "title": "1:1 Wendy Hernandez + Mia Brown",
-  "compiled_truth": "This one-on-one between [Wendy Hernandez](people/wendy-hernandez-80) and [Mia Brown](people/mia-brown-0) took place on January 1st, 2025—an unusual date for a meeting, suggesting some urgency or perhaps just aligning schedules over the holiday break. Both are associated with [Acme](companies/acme-0), the robotics company that's been making waves in warehouse automation.\n\nWendy came prepared with a list of concerns about the Q1 roadmap. She's been feeling stretched thin across multiple projects and wanted to discuss prioritization. Mia, as her manager, acknowledged that the team has been under significant pressure since the Series C closed. They spent roughly twenty minutes mapping out which deliverables could realistically slip versus which were tied to customer commitments.\n\nThe conversation shifted to career development around the halfway mark. Wendy expressed interest in taking on more technical leadership responsibilites, possibly moving toward a staff engineer track. Mia was supportive but noted that visibility matters—Wendy needs to present more at all-hands and write up her architectural decisions for broader consumption. They agreed to revisit this in their next 1:1 with concrete action items.\n\nThere was some discussion about team dynamics. One of the newer hires has been struggling with the codebase, and Wendy has been doing informal mentoring. Mia suggested formalizing this arrangement, which could count toward Wendy's leadership goals. They also touched on the upcoming robotics demo scheduled for late January—both are anxious about whether the grasping module will be ready.\n\nMia closed the meeting by checking in on work-life balance. The holiday timing of this meeting wasn't lost on her, and she encouraged Wendy to actually take some time off before the push begins. Wendy appreciated the sentiment but admited she'd probably be checking Slack anyway. The meeting ran about 45 minutes total, slightly over their usual thirty-minute slot.",
-  "timeline": "- **2024-06-15** | Wendy Hernandez joined [Acme](companies/acme-0) as Senior Engineer, reporting to Mia\n- **2024-07-22** | First formal 1:1 between Wendy and [Mia Brown](people/mia-brown-0) established bi-weekly cadence\n- **2024-09-10** | Wendy presented grasping module architecture to engineering team\n- **2024-11-03** | Mia nominated Wendy for internal technical excellence award\n- **2024-11-28** | Holiday schedules discussed; agreed to Jan 1 makeup meeting\n- **2025-01-01** | This 1:1 held—roadmap prioritization and career development discussed\n- **2025-01-15** | Follow-up scheduled to review staff engineer track progress\n- **2025-01-29** | Target date for robotics demo at [Acme](companies/acme-0) HQ",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-0-2025-01-01",
-    "name": "1:1 Wendy Hernandez + Mia Brown",
-    "meeting_type": "one_on_one",
-    "date": "2025-01-01",
-    "attendees": [
-      "people/wendy-hernandez-80",
-      "people/mia-brown-0"
-    ],
-    "topic_company": "companies/acme-0"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-1-2025-02-02.json b/eval/data/world-v1/meetings__oneonone-1-2025-02-02.json
deleted file mode 100644
index f55e7c352..000000000
--- a/eval/data/world-v1/meetings__oneonone-1-2025-02-02.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-1-2025-02-02",
-  "type": "meeting",
-  "title": "1:1 Carol Jackson + Victor Taylor",
-  "compiled_truth": "This one-on-one between [Carol Jackson](people/carol-jackson-81) and [Victor Taylor](people/victor-taylor-1) took place on February 2nd, 2025, focused on Beta's current security posture and upcoming product roadmap. Carol, as CTO, wanted Victor's perspective on the engineering team's capacity for the Q1 sprint.\n\nVictor opened with concerns about technical debt in the authentication module. He's been pushing for a refactor since November but keeps getting deprioritized. Carol acknowledged this but noted that [Beta](companies/beta-1)'s enterprise clients are demanding new compliance features faster than the team can ship them. They agreed to allocate two engineers to the auth refactor starting mid-February, with a hard deadline of March 15th.\n\nDiscussion moved to hiring. Victor mentioned he's been interviewing senior backend candidates but hasn't found anyone who meets Beta's bar for cryptography expertise. Carol suggested reaching out to her network from her previous role at Cloudflare—she knows a few people who might be intrested in the cybersecurity space. Victor seemed receptive but expressed worry about onboarding bandwidth.\n\nThe conversation got candid when Victor brought up team morale. Apparently there's been some friction between the platform and application teams over API ownership. Carol hadn't heard about this directly and asked Victor to set up a meeting with both tech leads so she can mediate. She emphasized that Beta can't afford internal conflicts when they're trying to close the Series B.\n\nThey briefly touched on the upcoming board presentation. Carol needs Victor to pull together metrics on system uptime and incident response times. Victor committed to having a draft by end of week. Carol mentioned she wants to highlight the zero-breach track record—it's a major selling point for enterprise prospects.\n\nMeeting wrapped with Carol asking about Victor's career goals. He expressed interest in eventually moving into a principal engineer role. Carol encouraged him to start mentoring junior devs more visibly and offered to sponsor him for the internal leadership program.",
-  "timeline": "- **2024-08-15** | Victor Taylor joined [Beta](companies/beta-1) as Senior Engineer on the platform team\n- **2024-10-22** | First 1:1 between Carol and Victor to discuss Q4 priorities\n- **2024-11-03** | Victor flagged auth module technical debt in team retrospective\n- **2024-12-10** | [Carol Jackson](people/carol-jackson-81) promoted Victor to lead the infrastructure workstream\n- **2025-01-14** | Beta's enterprise client audit revealed need for SOC 2 compliance updates\n- **2025-01-28** | Victor submitted proposal for authentication refactor project\n- **2025-02-02** | This 1:1 meeting held to align on Q1 roadmap and team concerns\n- **2025-02-09** | Follow-up meeting scheduled with platform and application tech leads",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-1-2025-02-02",
-    "name": "1:1 Carol Jackson + Victor Taylor",
-    "meeting_type": "one_on_one",
-    "date": "2025-02-02",
-    "attendees": [
-      "people/carol-jackson-81",
-      "people/victor-taylor-1"
-    ],
-    "topic_company": "companies/beta-1"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-10-2025-11-11.json b/eval/data/world-v1/meetings__oneonone-10-2025-11-11.json
deleted file mode 100644
index f1a612284..000000000
--- a/eval/data/world-v1/meetings__oneonone-10-2025-11-11.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-10-2025-11-11",
-  "type": "meeting",
-  "title": "1:1 Rosa Jackson + David Wang",
-  "compiled_truth": "This one-on-one between [Rosa Jackson](people/rosa-jackson-90) and [David Wang](people/david-wang-10) took place on November 11, 2025, as part of their regular sync cadence. Rosa, as CEO of [Beacon](companies/beacon-10), uses these sessions to stay connected with her technical leadership and get unfiltered reads on engineering morale and product velocity.\n\nDavid came prepared with concerns about the current sprint load. The team is stretched thin following the Q3 enterprise push, and he's seeing early signs of burnout among senior engineers. Two people have already asked about taking extended leave in December, which would put the v2.3 release at risk. Rosa acknowledged the pressure but pushed back gently—she needs the threat detection module shipped before the January renewal cycle with three major accounts.\n\nThey spent a good chunk of time discussing hiring. David wants to bring on two more backend engineers, ideally folks with experience in real-time data pipelines. Rosa agreed in principle but noted that the board is watching burn rate closely after the Series B. She suggested they revisit headcount after the Q4 close, assuming revenue targets are met. David seemed frustrated but understood teh constraints.\n\nConversation shifted to the competitive landscape. A new player, ClearShield, has been making noise on LinkedIn and reportedly poached someone from Palo Alto Networks. Rosa asked David to do some digging—she wants to understand their technical approach and whether they pose a real threat or are just marketing vapor. David promised to have a brief ready by end of week.\n\nThe meeting ended on a personal note. David mentioned his daughter's soccer tournament and Rosa asked about it genuinely. These small moments matter for retention, and Rosa knows David has options. She's been intentional about making sure he feels valued beyond just his output. They agreed to meet again in two weeks, with David owning the agenda for next time.",
-  "timeline": "- **2024-03-15** | Rosa Jackson hired David Wang as VP Engineering at Beacon\n- **2024-07-22** | First formal 1:1 structure established between Rosa and David\n- **2025-02-10** | David raised concerns about technical debt in 1:1; led to dedicated refactoring sprint\n- **2025-05-18** | Discussion about potential acquisition interest from CrowdStrike\n- **2025-08-30** | David flagged morale issues post-layoffs at competitor; Rosa approved retention bonuses\n- **2025-09-14** | 1:1 focused on Series B fundraising updates and board expectations\n- **2025-10-28** | Previous 1:1 cancelled due to Rosa's travel to NYC investor meetings\n- **2025-11-11** | Current meeting: discussed burnout, hiring constraints, competitive threats from ClearShield",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-10-2025-11-11",
-    "name": "1:1 Rosa Jackson + David Wang",
-    "meeting_type": "one_on_one",
-    "date": "2025-11-11",
-    "attendees": [
-      "people/rosa-jackson-90",
-      "people/david-wang-10"
-    ],
-    "topic_company": "companies/beacon-10"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-11-2025-12-12.json b/eval/data/world-v1/meetings__oneonone-11-2025-12-12.json
deleted file mode 100644
index a2d53bb09..000000000
--- a/eval/data/world-v1/meetings__oneonone-11-2025-12-12.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-11-2025-12-12",
-  "type": "meeting",
-  "title": "1:1 Chris Jackson + Mark Thomas",
-  "compiled_truth": "One-on-one meeting between [Chris Jackson](people/chris-jackson-91) and [Mark Thomas](people/mark-thomas-11) held on December 12th, 2025. Chris is the CEO of [Compass](companies/compass-11), a crypto infrastructure company focused on institutional-grade custody solutions. Mark Thomas serves as VP of Engineering at the same company.\n\nThe meeting opened with Mark raising concerns about the current sprint velocity. The engineering team has been stretched thin following the departure of two senior backend engineers last month. Chris acknowledged the hiring gap and confirmed that three offers were extended this week, with two candidates already in final negotiations. Should close by end of year if compensation benchmarks hold.\n\nDiscussion moved to the upcoming SOC 2 Type II audit scheduled for Q1. Mark flagged that the security logging infrastructure needs attention before auditors arrive. Estimated two weeks of dedicated work from the platform team. Chris agreed to deprioritize the mobile wallet feature to free up resources—compliance comes first when you're courting institutional clients.\n\nMark also brought up team morale issues. The recent reorg created some confusion around reporting structures, particularly for the DevOps folks who now report into platform instead of infrastructure. Chris suggested scheduling a team all-hands to clarify the reasoning and address questions directly. Transparency matters here.\n\nOn the product side, Chris shared early feedback from the Fidelity pilot program. Their team is impressed with latency numbers but wants better documentation around the API authentication flow. Mark committed to having technical writing resources assigned by Monday. The partnership could be transformative if executed well—Fidelity's stamp of approval opens doors across traditional finance.\n\nFinally, they discussed Mark's own career trajectory. He's been in the VP role for eighteen months now and expressed interest in eventually moving toward a CTO track. Chris was supportive, noting that the company's growth trajectory should create opportunites for expanded scope. They agreed to revisit this conversation in Q2 after the Series B closes.",
-  "timeline": "- **2024-03-15** | Mark Thomas promoted to VP of Engineering at [Compass](companies/compass-11)\n- **2024-06-22** | Chris and Mark present custody architecture at Consensus conference\n- **2024-09-10** | Engineering team expanded to 28 people under Mark's leadership\n- **2025-02-14** | SOC 2 Type I certification achieved after 6-month preparation\n- **2025-05-30** | [Chris Jackson](people/chris-jackson-91) announces Fidelity pilot partnership internally\n- **2025-08-18** | Team reorg consolidates DevOps and platform engineering\n- **2025-10-02** | Two senior engineers depart for competitor roles\n- **2025-11-20** | Previous 1:1 focused on hiring pipeline and Q4 priorities\n- **2025-12-12** | This meeting: discussed hiring, SOC 2 prep, team morale, Fidelity feedback",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-11-2025-12-12",
-    "name": "1:1 Chris Jackson + Mark Thomas",
-    "meeting_type": "one_on_one",
-    "date": "2025-12-12",
-    "attendees": [
-      "people/chris-jackson-91",
-      "people/mark-thomas-11"
-    ],
-    "topic_company": "companies/compass-11"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-12-2025-01-13.json b/eval/data/world-v1/meetings__oneonone-12-2025-01-13.json
deleted file mode 100644
index 21a636144..000000000
--- a/eval/data/world-v1/meetings__oneonone-12-2025-01-13.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-12-2025-01-13",
-  "type": "meeting",
-  "title": "1:1 Sarah Williams + Henry Johnson",
-  "compiled_truth": "One-on-one sync between [Sarah Williams](people/sarah-williams-92) and [Henry Johnson](people/henry-johnson-12) held on January 13th, 2025. This was a scheduled check-in to discuss ongoing work at [Lumen](companies/lumen-12) and address some operational concerns that had been building up over the past few weeks.\n\nSarah came into the meeting wanting to talk about team bandwidth. She's been feeling stretched thin since the Q4 push and mentioned that the current sprint cadence isn't sustainable. Henry acknowledged this and said he'd been hearing similar concerns from other leads. They agreed to revisit the sprint structure at the next all-hands, possibly moving to a more flexible model that accounts for research timelines better.\n\nHenry brought up the upcoming Series B prep work. He wants Sarah more involved in the investor materials since she has the deepest understanding of the platform's technical differentiation. She was receptive but flagged that she'd need to offload some of her current project oversight to make room. They discussed potentially bringing in a senior hire to backfil some of her responsibilities — someone with biotech ops experience ideally.\n\nConversation drifted a bit toward the recent FDA guidance changes affecting Lumen's regulatory pathway. Neither of them had fully digested the implications yet, but Henry mentioned he'd scheduled a call with outside counsel for later in the week. Sarah offered to loop in the science team to prepare questions.\n\nMood was generally positive despite the workload concerns. Henry expressed appreciation for how Sarah handled the December crunch and mentioned she's being considered for an expanded role once the Series B closes. Sarah seemed pleased but didn't commit to anything — said she wanted to see how the next quarter shapes up first.\n\nAction items: Sarah to draft bandwidth proposal by EOW. Henry to share counsel call notes. Both to sync again after the all-hands on sprint restructuring.",
-  "timeline": "- **2024-09-15** | Sarah Williams joins [Lumen](companies/lumen-12) as VP of Platform Operations\n- **2024-10-22** | First 1:1 between Sarah and [Henry Johnson](people/henry-johnson-12) to establish working relationship\n- **2024-11-08** | Sarah leads internal review of Q4 deliverables timeline\n- **2024-12-02** | Henry asks Sarah to take point on December platform push\n- **2024-12-19** | Successful completion of Q4 milestones, team exhausted but deliverables met\n- **2025-01-06** | Sarah flags bandwidth concerns via Slack ahead of scheduled 1:1\n- **2025-01-13** | This meeting — discussed sprint cadence, Series B prep, potential senior hire\n- **2025-01-20** | Planned all-hands to address sprint restructuring",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-12-2025-01-13",
-    "name": "1:1 Sarah Williams + Henry Johnson",
-    "meeting_type": "one_on_one",
-    "date": "2025-01-13",
-    "attendees": [
-      "people/sarah-williams-92",
-      "people/henry-johnson-12"
-    ],
-    "topic_company": "companies/lumen-12"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-13-2025-02-14.json b/eval/data/world-v1/meetings__oneonone-13-2025-02-14.json
deleted file mode 100644
index 846f5c919..000000000
--- a/eval/data/world-v1/meetings__oneonone-13-2025-02-14.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-13-2025-02-14",
-  "type": "meeting",
-  "title": "1:1 Eric Martinez + Mia Lee – February 2025",
-  "compiled_truth": "Regular one-on-one sync between [Eric Martinez](people/eric-martinez-93) and [Mia Lee](people/mia-lee-13), held mid-February 2025 to discuss ongoing priorities and team dynamics at [Cipher](companies/cipher-13). Eric scheduled this as part of his bi-weekly cadence with direct reports on the product side.\n\nMia came prepared with updates on the fraud detection module rollout, which had hit a minor snag with false positive rates climbing higher than expected. She walked Eric through the data—roughly 3.2% of legitimate transactions getting flagged, up from the target 1.8%. Eric pushed back gently, asking whether the ML team had been looped in early enough or if this was a case of shipping too fast. Mia acknowledged the timeline pressure but defended the decision to launch, noting that waiting another sprint would've meant missing the Q1 commitment to enterprise clients.\n\nConversation shifted to team morale. Mia mentioned that two engineers on her squad were showing signs of burnout after the holiday crunch. She proposed rotating them onto lower-intensity projects for a few weeks. Eric agreed but flagged that headcount constraints meant backfills weren't coming until Q2 at the earliest. They discussed whether contracting out some QA work could relieve presure.\n\nEric also brought up the topic of Mia's career trajectory. He'd been impressed with how she handled the Meridian Bank integration last quarter and wanted to know if she was interested in stepping into a senior PM role. Mia seemed receptive but expressed concern about losing hands-on time with the product. Eric suggested she shadow the VP of Product for a few sessions to get a feel for what the next level looks like.\n\nAction items: Mia to draft a burnout mitigation plan by EOW, Eric to connect her with recruiting about contractor options, and both to revisit the promotion conversation in their March sync.",
-  "timeline": "- **2024-09-18** | Mia Lee joins [Cipher](companies/cipher-13) as Product Manager, reporting to Eric\n- **2024-10-30** | First formal 1:1 between Eric and Mia to establish working norms\n- **2024-12-05** | Mia leads Meridian Bank integration launch, earning recognition from leadership\n- **2025-01-10** | Eric flags Mia as high-potential in quarterly talent review\n- **2025-01-24** | Previous 1:1 focused on Q1 roadmap prioritization\n- **2025-02-14** | This meeting—discussed fraud module issues, team burnout, promotion path\n- **2025-02-21** | Mia scheduled to present burnout mitigation proposal to Eric\n- **2025-03-14** | Next 1:1 planned to revisit senior PM conversation",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-13-2025-02-14",
-    "name": "1:1 Eric Martinez + Mia Lee",
-    "meeting_type": "one_on_one",
-    "date": "2025-02-14",
-    "attendees": [
-      "people/eric-martinez-93",
-      "people/mia-lee-13"
-    ],
-    "topic_company": "companies/cipher-13"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-14-2025-03-15.json b/eval/data/world-v1/meetings__oneonone-14-2025-03-15.json
deleted file mode 100644
index 7198c82e9..000000000
--- a/eval/data/world-v1/meetings__oneonone-14-2025-03-15.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-14-2025-03-15",
-  "type": "meeting",
-  "title": "1:1 Rosa Nakamura + Vera Chen",
-  "compiled_truth": "This one-on-one between [Rosa Nakamura](people/rosa-nakamura-94) and [Vera Chen](people/vera-chen-14) covered several threads related to Mosaic's growth trajectory and Rosa's expanding role on the team. Rosa has been with [Mosaic](companies/mosaic-14) for about eight months now, having joined as a senior product manager focused on creator tools.\n\nVera opened by checking in on Rosa's bandwidth given the recent push to ship the new collab features. Rosa mentioned feeling stretched thin but not burned out—she's been leaning on her PM counterpart on the discovery team to share some of the load. The two discussed whether it makes sense to bring on a more junior PM to support Rosa directly, or whether the current structure can hold through Q2. Vera seemed open to accelerating the hire if Rosa feels strongly about it.\n\nConversation shifted to Rosa's longer-term interests. She expressed curiosity about eventually moving into a more strategic role, possibly overseeing multiple product areas rather than just creator-facing features. Vera noted that this aligns with how she sees Rosa's trajectory at the company, though she cautioned that it will depend on how the next funding round shapes headcount plans. Rosa apreciated the transparency.\n\nThey also touched on some interpersonal dynamics. Rosa flagged that there's been some friction between eng and design on the collab feature—nothing major, but worth monitoring. Vera asked Rosa to keep her posted if it escalates, and offered to step in if needed.\n\nFinally, Rosa brought up the upcoming offsite. She's excited about it but had some logistical questions about travel reimbursement that Vera promised to clarify with ops. Meeting wrapped with Vera encouraging Rosa to take a day off after the collab launch—she's earned it.",
-  "timeline": "- **2024-07-08** | Rosa Nakamura joins Mosaic as Senior PM, Creator Tools\n- **2024-09-22** | First 1:1 between Rosa and Vera to establish working cadence\n- **2024-11-15** | Rosa presents creator monetization roadmap to leadership team\n- **2025-01-10** | Collab feature development kicks off with Rosa as PM lead\n- **2025-02-28** | Rosa raises concerns about eng/design alignment in slack DM to Vera\n- **2025-03-15** | This 1:1 covering workload, career growth, and team dynamics\n- **2025-03-20** | Collab feature scheduled for internal beta launch",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-14-2025-03-15",
-    "name": "1:1 Rosa Nakamura + Vera Chen",
-    "meeting_type": "one_on_one",
-    "date": "2025-03-15",
-    "attendees": [
-      "people/rosa-nakamura-94",
-      "people/vera-chen-14"
-    ],
-    "topic_company": "companies/mosaic-14"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-15-2025-04-16.json b/eval/data/world-v1/meetings__oneonone-15-2025-04-16.json
deleted file mode 100644
index d340dfa55..000000000
--- a/eval/data/world-v1/meetings__oneonone-15-2025-04-16.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-15-2025-04-16",
-  "type": "meeting",
-  "title": "1:1 Rachel Brown + Noah Kapoor",
-  "compiled_truth": "This one-on-one between [Rachel Brown](people/rachel-brown-95) and [Noah Kapoor](people/noah-kapoor-15) took place on April 16, 2025, as part of their regular sync cadence. Rachel, who leads product at [Tessera](companies/tessera-15), has been meeting with Noah bi-weekly since he joined the company's data infrastructure team last fall.\n\nThe conversation opened with Noah raising concerns about the current sprint velocity. He feels the team is stretched thin across too many initiatives, particularly with the new real-time fraud detection module that's been prioritized for Q2. Rachel acknowledged the pressure but pushed back slightly—she noted that the fraud work is non-negotiable given recent chargebacks hitting enterprise clients. They agreed to revisit scope on the analytics dashboard refresh, potentially pushing some features to Q3.\n\nNoah also brought up a interpersonal issue with one of the senior engineers on his squad. Apparently there's been some tension around code review standards, with the senior engineer blocking PRs for what Noah considers minor style issues. Rachel offered to facilitate a conversation but encouraged Noah to try addressing it directly first. She shared a framework she's used before: lead with curiousity, assume good intent, propose concrete changes.\n\nOn career development, Noah expressed interest in eventually moving toward a tech lead role. Rachel was supportive and suggested he start by taking ownership of the upcoming data pipeline migration project. \"It's meaty enough to demonstrate leadership but scoped enough that you wont drown,\" she said. They discussed what success would look like and agreed to revisit the topic in their June check-in.\n\nRachel closed by mentioning that [Tessera](companies/tessera-15) is planning an offsite in early May and asked Noah to think about topics for a team retrospective session. She also reminded him about the upcoming all-hands where the CEO will be sharing updated roadmap priorities. Overall a productive session with clear action items on both sides.",
-  "timeline": "- **2024-09-23** | [Noah Kapoor](people/noah-kapoor-15) joins Tessera's data infrastructure team\n- **2024-10-14** | First 1:1 between Rachel and Noah established as bi-weekly sync\n- **2024-12-05** | Noah completes onboarding project, praised in team retro\n- **2025-01-22** | Rachel and Noah discuss Q1 goals, fraud detection work first mentioned\n- **2025-03-10** | Sprint planning meeting flags resource constraints on Noah's squad\n- **2025-04-16** | This 1:1 takes place, covering velocity, team dynamics, and career growth\n- **2025-05-08** | Planned Tessera team offsite in Lake Tahoe\n- **2025-06-01** | Scheduled follow-up on Noah's tech lead trajectory",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-15-2025-04-16",
-    "name": "1:1 Rachel Brown + Noah Kapoor",
-    "meeting_type": "one_on_one",
-    "date": "2025-04-16",
-    "attendees": [
-      "people/rachel-brown-95",
-      "people/noah-kapoor-15"
-    ],
-    "topic_company": "companies/tessera-15"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-16-2025-05-17.json b/eval/data/world-v1/meetings__oneonone-16-2025-05-17.json
deleted file mode 100644
index 222e98e0f..000000000
--- a/eval/data/world-v1/meetings__oneonone-16-2025-05-17.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-16-2025-05-17",
-  "type": "meeting",
-  "title": "1:1 Chris Singh + Ulrich Wang",
-  "compiled_truth": "One-on-one meeting between [Chris Singh](people/chris-singh-96) and [Ulrich Wang](people/ulrich-wang-16) to discuss Ulrich's work at [Mantle](companies/mantle-16) and broader career trajectory. Chris has been tracking Mantle since their seed round and wanted to get Ulrich's on-the-ground perspective after six months at the company.\n\nUlrich opened with an update on Mantle's recent product pivot. The team has shifted away from their original audio-first social concept toward something more visual—think collaborative mood boards with social graph mechanics. He mentioned the pivot came after disappointing retention numbers in Q1, though he seemed optimistic about early signals from the new direction. DAUs on the beta have been climbing steadily, mostly concentrated in college-age users on the West Coast.\n\nDiscussion moved to team dynamics. Ulrich noted that the engineering org has grown to about 18 people now, with three new hires starting in April. He's been leading a small pod focused on the discovery algorithm, which he described as \"the most interesting problem I've worked on.\" There's apparently some tension between product and engineering leadership about pacing—the CEO wants to ship faster, but Ulrich thinks they're already pushing the limits of what's sustainable.\n\nChris asked about fundraising outlook. Ulrich said there's been inbound interest from a few Series A firms, though nothing concrete. The company has about 14 months of runway left at current burn. Mantle's leadership seems focused on hitting certain engagement milestones before going out formally, probably late summer or early fall.\n\nToward the end, conversation shifted to Ulrich's personal goals. He's been at Mantle since late 2024 and is still learning a ton, but he's starting to think about what a founding role might look like down the line. Chris offerd to introduce him to a few founders who made similar transitions. They agreed to reconnect in July to see how things have progressed.",
-  "timeline": "- **2024-11-04** | [Ulrich Wang](people/ulrich-wang-16) joins Mantle as senior engineer on growth team\n- **2025-01-22** | Mantle leadership decides to pivot away from audio-first concept\n- **2025-03-10** | Ulrich begins leading discovery algorithm pod at Mantle\n- **2025-04-15** | Three new engineering hires onboard at [Mantle](companies/mantle-16)\n- **2025-05-17** | 1:1 meeting between [Chris Singh](people/chris-singh-96) and Ulrich Wang\n- **2025-05-18** | Chris sends follow-up email with founder intro suggestions\n- **2025-07-20** | Planned follow-up conversation scheduled",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-16-2025-05-17",
-    "name": "1:1 Chris Singh + Ulrich Wang",
-    "meeting_type": "one_on_one",
-    "date": "2025-05-17",
-    "attendees": [
-      "people/chris-singh-96",
-      "people/ulrich-wang-16"
-    ],
-    "topic_company": "companies/mantle-16"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-17-2025-06-18.json b/eval/data/world-v1/meetings__oneonone-17-2025-06-18.json
deleted file mode 100644
index 616328476..000000000
--- a/eval/data/world-v1/meetings__oneonone-17-2025-06-18.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-17-2025-06-18",
-  "type": "meeting",
-  "title": "1:1 Tina Hernandez + Quinten Wang",
-  "compiled_truth": "Bi-weekly sync between [Tina Hernandez](people/tina-hernandez-97) and [Quinten Wang](people/quinten-wang-17) to discuss [Gravity](companies/gravity-17) commercial strategy and Quinten's transition into a more client-facing role. This meeting has become increasingly important as Gravity scales its enterprise partnerships.\n\nTina opened by checking in on Quinten's workload after the Series B close. He mentioned feeling stretched thin across three major accounts but said the new hire starting next month should help. They discussed the upcoming Roche pilot—Quinten flagged some concerns about timeline slippage on the data integration side. Tina suggested looping in the engineering lead earlier rather than waiting for the formal handoff.\n\nConversation shifted to Quinten's career development. He's been with Gravity for two years now and expressed interest in eventually moving toward a BD leadership track. Tina was supportive, noting that his technical background gives him an edge in pharma conversations that most sales folks don't have. She recommended he start shadowing some of her board prep meetings to get exposure to the strategic layer.\n\nThey spent a few minutes on the Novartis renewal coming up in Q3. Pricing discussions have been tricky—Novartis is pushing for a multi-year discount that Gravity's finance team isn't thrilled about. Quinten thinks there's room to negotiate on scope instead of pure price reduction. Tina agreed to bring this framing to the exec team.\n\nQuinten also brought up some friction with the product team around feature requests from his accounts. He feels like enterprise feedback isn't being prioritized. Tina acknowledged this is a company-wide tension right now and said she's working on a better intake process with the CPO. No immediate resolution but she appreicates him raising it.\n\nMeeting wrapped with a quick discussion of the JP Morgan Healthcare Conference in January. Tina wants Quinten to attend this year and help staff the Gravity booth. He's excited about the opportunity.",
-  "timeline": "- **2023-08-14** | Quinten Wang joins Gravity as Senior Account Manager, reports to Tina Hernandez\n- **2024-02-22** | First major enterprise deal closed by Quinten — Merck pilot program\n- **2024-06-10** | Tina promotes Quinten to lead the Novartis relationship after strong Q1 performance\n- **2024-11-05** | 1:1 discussion about Quinten's long-term career goals at [Gravity](companies/gravity-17)\n- **2025-03-18** | [Tina Hernandez](people/tina-hernandez-97) gives Quinten stretch assignment on Series B investor materials\n- **2025-04-30** | Quinten presents at internal all-hands on enterprise account learnings\n- **2025-06-18** | Current 1:1 sync covering Roche pilot, Novartis renewal, and career development\n- **2025-09-15** | Planned: Quinten to present Q3 enterprise metrics to board",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-17-2025-06-18",
-    "name": "1:1 Tina Hernandez + Quinten Wang",
-    "meeting_type": "one_on_one",
-    "date": "2025-06-18",
-    "attendees": [
-      "people/tina-hernandez-97",
-      "people/quinten-wang-17"
-    ],
-    "topic_company": "companies/gravity-17"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-18-2025-07-19.json b/eval/data/world-v1/meetings__oneonone-18-2025-07-19.json
deleted file mode 100644
index 43b430baf..000000000
--- a/eval/data/world-v1/meetings__oneonone-18-2025-07-19.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-18-2025-07-19",
-  "type": "meeting",
-  "title": "1:1 Rosa Miller + Nina Rodriguez",
-  "compiled_truth": "One-on-one meeting between [Rosa Miller](people/rosa-miller-98) and [Nina Rodriguez](people/nina-rodriguez-18) held on July 19, 2025. Rosa serves as VP of Engineering at [Apex](companies/apex-18), while Nina reports to her as a senior infrastructure engineer on the GPU cluster team.\n\nThe meeting opened with Nina raising concerns about the current on-call rotation burnout affecting her squad. Three engineers have been handling overnight pages for the past six weeks due to staffing gaps, and Nina's worried about attrition if they don't address it soon. Rosa acknowledged the issue and committed to escalating headcount requests to the exec team before end of quarter.\n\nDiscussion shifted to Nina's career trajectory. She's been at Apex for almost three years now and expressed interest in moving toward a tech lead role. Rosa was supportive, noting that Nina's work on the fault-tolerant scheduling system has been exactly the kind of technical leadership they need more of. They agreed to identify a meaty project for Q4 that would give Nina more visibility with cross-functional partners.\n\nThere was a brief tangent about the upcoming datacenter expansion in Austin. Nina mentioned some architectural concerns with how they're planning to handle network partitions—she thinks the current design is too optimistic about latency assumptions. Rosa asked her to write up a short doc outlining the risks so they can bring it to the infrastructure review meeting next week.\n\nThe tone throughout was candid and collaborative. Nina seems genuinly engaged with the technical challenges but is clearly feeling the strain of being understaffed. Rosa's management style came across as hands-on; she asked detailed questions about specific systems rather than staying at a high level. Meeting wrapped early at about 40 minutes. Next 1:1 scheduled for two weeks out.",
-  "timeline": "- **2022-09-12** | Nina Rodriguez joins [Apex](companies/apex-18) as infrastructure engineer\n- **2023-03-15** | Rosa Miller promoted to VP of Engineering at Apex\n- **2023-11-20** | Nina takes over GPU cluster reliability from departing tech lead\n- **2024-02-08** | First 1:1 between Rosa and Nina after reorg\n- **2024-07-22** | Nina presents fault-tolerant scheduler design to infrastructure team\n- **2024-12-03** | Rosa nominates Nina for internal engineering excellence award\n- **2025-04-14** | Discussion about Nina's tech lead aspirations begins\n- **2025-07-19** | This 1:1 meeting covering on-call burnout and career development",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-18-2025-07-19",
-    "name": "1:1 Rosa Miller + Nina Rodriguez",
-    "meeting_type": "one_on_one",
-    "date": "2025-07-19",
-    "attendees": [
-      "people/rosa-miller-98",
-      "people/nina-rodriguez-18"
-    ],
-    "topic_company": "companies/apex-18"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-19-2025-08-20.json b/eval/data/world-v1/meetings__oneonone-19-2025-08-20.json
deleted file mode 100644
index 8b9211899..000000000
--- a/eval/data/world-v1/meetings__oneonone-19-2025-08-20.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-19-2025-08-20",
-  "type": "meeting",
-  "title": "1:1 Kate Lopez + Adam Lee",
-  "compiled_truth": "One-on-one meeting between [Kate Lopez](people/kate-lopez-99) and [Adam Lee](people/adam-lee-19) to discuss Forge's current trajectory and Adam's role within the organization. Kate, as a board member and early investor, wanted to get a direct read on engineering morale following the recent reorg that consolidated the wallet and exchange teams.\n\nAdam opened by walking through the technical debt situation. He's been pushing to allocate 20% of sprint capacity to infrastructure work, but product keeps overriding with feature requests. The new custody solution is behind schedule—partly because two senior engineers left in Q2 and backfills have been slow. Kate asked pointed questions about retention. Adam was candid: comp is competative but the equity refresh cycle hasn't kept pace with market. People are getting poached by better-funded L2 projects.\n\nDiscussion turned to [Forge](companies/forge-19)'s positioning in the institutional custody market. Adam believes the technical moat is real but shrinking. Competitors are catching up on the multi-sig implementation that was once a differentiator. He advocated for doubling down on the API layer—make Forge the infrastructure other exchanges build on rather than competing directly for retail flow.\n\nKate shared some board-level context. Series C conversations are progressing but investors are scrutinizing the path to profitability more aggressively than in previous rounds. She hinted that some board members want to see headcount cuts, which Adam pushed back on strongly. Cutting eng now would delay the enterprise product by at least two quarters.\n\nThey also touched on Adam's career trajectory. He's been leading the platform team for 18 months and is starting to feel the itch for something new. Kate floated the idea of him taking on a broader architecture role, maybe even a CTO track if current leadership changes. Adam seemed receptive but non-committal.\n\nAction items: Adam to put together a retention proposal with specific equity recommendations. Kate to advocate for protecting eng headcount in upcoming board meeting. Follow-up scheduled for early September.",
-  "timeline": "- **2024-03-15** | [Adam Lee](people/adam-lee-19) promoted to platform team lead at Forge\n- **2024-07-22** | Kate Lopez joins Forge board following Series B close\n- **2024-11-08** | First 1:1 between Kate and Adam to discuss eng culture\n- **2025-02-14** | Adam presents infrastructure roadmap to board, receives approval for dedicated DevOps hire\n- **2025-05-19** | Reorg consolidates wallet and exchange teams under unified leadership\n- **2025-06-30** | Two senior engineers depart [Forge](companies/forge-19) for competing L2 projects\n- **2025-08-20** | This meeting: retention concerns, Series C dynamics, Adam's career path\n- **2025-09-04** | Follow-up meeting scheduled to review retention proposal",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-19-2025-08-20",
-    "name": "1:1 Kate Lopez + Adam Lee",
-    "meeting_type": "one_on_one",
-    "date": "2025-08-20",
-    "attendees": [
-      "people/kate-lopez-99",
-      "people/adam-lee-19"
-    ],
-    "topic_company": "companies/forge-19"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-2-2025-03-03.json b/eval/data/world-v1/meetings__oneonone-2-2025-03-03.json
deleted file mode 100644
index 7bcdded98..000000000
--- a/eval/data/world-v1/meetings__oneonone-2-2025-03-03.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-2-2025-03-03",
-  "type": "meeting",
-  "title": "1:1 Iris Lee + Mark Jones",
-  "compiled_truth": "This one-on-one between [Iris Lee](people/iris-lee-82) and [Mark Jones](people/mark-jones-2) took place on March 3rd, 2025, primarily to discuss ongoing work at [Gamma](companies/gamma-2) and Mark's career trajectory within the fintech org. Iris, as Mark's direct manager on the platform team, scheduled this as part of her regular cadence with senior engineers.\n\nMark came prepared with a few items. First up was the status of the payment reconciliation service migration—he's about 70% through refactoring the legacy batch processor into the new event-driven architecture. Some concerns about test coverage in the edge cases around failed ACH transfers. Iris pushed back a bit, asking whether the team had documented the failure modes properly. Mark admitted the runbooks were outdated and committed to updating them before the next deploy window.\n\nConversation shifted to Mark's interest in taking on more technical leadership. He's been at Gamma for nearly two years now and feels ready to mentor some of the newer engineers on the team. Iris was supportive but noted that the promo cycle doesn't start until Q3, so they'd need to build a case over the next few months. She suggested he lead the upcoming API versioning initiative as a way to demonstrate cross-team coordination skills.\n\nThey also touched on team dynamics briefly. There's been some tension between the platform and compliance teams over audit logging requirements—Mark mentioned feeling like the compliance folks don't understand the performance implications of their requests. Iris acknowledged the friction and said she'd bring it up with leadership. She reminded Mark to keep communication constructive in Slack threads.\n\nAction items: Mark to finish reconciliation migration by end of March, update the runbooks, and draft a short proposal for leading the API versioning project. Iris to sync with compliance team lead about the logging debate. Next 1:1 scheduled for March 17th.",
-  "timeline": "- **2023-06-12** | Mark Jones joins [Gamma](companies/gamma-2) as Senior Software Engineer on platform team\n- **2023-09-08** | Iris Lee becomes Mark's manager after reorg\n- **2024-02-14** | First major project shipped: real-time balance API\n- **2024-07-22** | Mark presents at internal tech talk on event sourcing patterns\n- **2024-11-03** | Payment reconciliation migration project kicks off\n- **2025-01-20** | [Iris Lee](people/iris-lee-82) flags Mark as potential tech lead candidate in planning docs\n- **2025-03-03** | 1:1 meeting to discuss migration progress and career growth\n- **2025-03-17** | Follow-up 1:1 scheduled",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-2-2025-03-03",
-    "name": "1:1 Iris Lee + Mark Jones",
-    "meeting_type": "one_on_one",
-    "date": "2025-03-03",
-    "attendees": [
-      "people/iris-lee-82",
-      "people/mark-jones-2"
-    ],
-    "topic_company": "companies/gamma-2"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-20-2025-09-21.json b/eval/data/world-v1/meetings__oneonone-20-2025-09-21.json
deleted file mode 100644
index e316d613e..000000000
--- a/eval/data/world-v1/meetings__oneonone-20-2025-09-21.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-20-2025-09-21",
-  "type": "meeting",
-  "title": "1:1 Zoe Gonzalez + Vera Singh",
-  "compiled_truth": "One-on-one meeting between [Zoe Gonzalez](people/zoe-gonzalez-100) and [Vera Singh](people/vera-singh-20) to discuss Kindle's current trajectory and Vera's role on the growth team. Zoe initiated this meeting after noticing some friction in cross-functional comms between growth and product.\n\nVera came prepared with a list of blockers she's been experiencing, primarily around data access and the lag time getting experiment results from the analytics team. She mentioned feeling like growth initatives keep getting deprioritized when engineering resources get tight. Zoe acknowledged this has been an ongoing tension and committed to raising it in the next leadership sync.\n\nThey spent a good chunk of time talking about [Kindle](companies/kindle-20)'s upcoming Series B prep and how the growth metrics will factor into the narrative. Vera expressed some anxiety about whether her team's work would be properly attributed in the fundraise materials. Zoe reassured her that the investor deck specifically calls out the 3x improvement in activation rates that Vera's experiments drove.\n\nConversation shifted to career development. Vera mentioned she's been at Kindle for almost two years now and is starting to think about what's next for her professionally. She's interested in eventually moving into a product management role, possibly on the core product side rather than growth. Zoe suggested she start shadowing some PM meetings and offered to connect her with a few PMs she knows who made similar transitions.\n\nThey briefly touched on team morale. Vera noted that her direct reports have been feeling the crunch of aggressive targets. Zoe asked her to keep an eye on burnout signals and reminded her that sustainable pace matters more than hitting every weekly goal. The meeting wrapped with Vera agreeing to send over a written summary of her top three asks for the leadership team.",
-  "timeline": "- **2023-11-14** | Vera Singh joins [Kindle](companies/kindle-20) as Growth Lead, reporting to Zoe\n- **2024-03-22** | First formal 1:1 between Zoe and Vera to establish working cadence\n- **2024-08-09** | Vera presents activation experiment results at all-hands, praised by leadership\n- **2024-12-01** | [Zoe Gonzalez](people/zoe-gonzalez-100) promotes Vera to Senior Growth Lead\n- **2025-04-17** | Tensions surface between growth and product teams over resourcing\n- **2025-07-30** | Vera flags burnout concerns on her team to Zoe via Slack\n- **2025-09-21** | This 1:1 meeting takes place to address blockers and career growth",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-20-2025-09-21",
-    "name": "1:1 Zoe Gonzalez + Vera Singh",
-    "meeting_type": "one_on_one",
-    "date": "2025-09-21",
-    "attendees": [
-      "people/zoe-gonzalez-100",
-      "people/vera-singh-20"
-    ],
-    "topic_company": "companies/kindle-20"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-21-2025-10-22.json b/eval/data/world-v1/meetings__oneonone-21-2025-10-22.json
deleted file mode 100644
index 5c062c34c..000000000
--- a/eval/data/world-v1/meetings__oneonone-21-2025-10-22.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-21-2025-10-22",
-  "type": "meeting",
-  "title": "1:1 Chris Miller + Eric Lee",
-  "compiled_truth": "This one-on-one between [Chris Miller](people/chris-miller-101) and [Eric Lee](people/eric-lee-21) covered ground on Lucid's Series B positioning and Eric's evolving role as the company scales. Chris opened by checking in on Eric's bandwidth—he'd noticed Eric pulling longer hours since the carbon accounting feature shipped last month.\n\nEric shared that he's been stretched thin managing both the data engineering team and serving as the de facto point person for enterprise integrations. The [Lucid](companies/lucid-21) platform has seen a 3x increase in API calls from manufacturing clients, which is great for growth but creating technical debt that's starting to pile up. Chris pushed on whether Eric needs another senior hire or if the issue is more about process. Eric leaned toward the latter, suggesting they need better runbooks before adding headcount.\n\nThey spent a good chunk of time discussing the upcoming board meeting. Chris wants Eric to present the technical roadmap slide, specifically the emissions data pipeline improvements that have reduced customer onboarding time from 6 weeks to 11 days. Eric was hesitant—he's not super comfortable with board-level presentations—but Chris framed it as a growth opportunity and committed to doing a practice run with him beforehand.\n\nConversation shifted to team dynamics. Eric mentioned some tension between the frontend and backend teams around ownership of the dashboard performance issues. Chris suggested Eric facilitate a blameless postmortem, noting that these frictions tend to resolve themselves when you get people in a room with a whiteboard. Eric agreed to schedule somthing for next week.\n\nThey briefly touched on Eric's career trajectory. He's been at Lucid for almost three years now and is starting to think about what VP-level responsibilties would look like. Chris was encouraging but honest: the company needs to see Eric operating more strategically and less in the weeds before making that jump. They agreed to revisit this in Q1 after the Series B closes.\n\nAction items: Eric to draft postmortem agenda by Friday, Chris to send board deck template, and they'll sync again before the November board meeting.",
-  "timeline": "- **2022-04-15** | Eric Lee joins [Lucid](companies/lucid-21) as Senior Data Engineer, first infrastructure hire\n- **2023-02-10** | Eric promoted to Lead Engineer after shipping core emissions tracking pipeline\n- **2023-09-22** | First 1:1 between [Chris Miller](people/chris-miller-101) and Eric to discuss team expansion\n- **2024-03-18** | Eric presents technical architecture at Lucid's Series A close celebration\n- **2024-08-05** | Data engineering team grows to 4 reports under Eric's leadership\n- **2024-11-30** | Eric leads migration to new cloud provider, completes under budget\n- **2025-06-12** | Carbon accounting feature ships, Eric credited as technical lead\n- **2025-10-22** | This 1:1 discussing Series B prep and Eric's career development\n- **2025-11-15** | Scheduled board meeting where Eric will present roadmap slide",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-21-2025-10-22",
-    "name": "1:1 Chris Miller + Eric Lee",
-    "meeting_type": "one_on_one",
-    "date": "2025-10-22",
-    "attendees": [
-      "people/chris-miller-101",
-      "people/eric-lee-21"
-    ],
-    "topic_company": "companies/lucid-21"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-22-2025-11-23.json b/eval/data/world-v1/meetings__oneonone-22-2025-11-23.json
deleted file mode 100644
index 7fe3c1a8f..000000000
--- a/eval/data/world-v1/meetings__oneonone-22-2025-11-23.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-22-2025-11-23",
-  "type": "meeting",
-  "title": "1:1 Kevin Taylor + Quinten Rodriguez",
-  "compiled_truth": "This one-on-one between [Kevin Taylor](people/kevin-taylor-102) and [Quinten Rodriguez](people/quinten-rodriguez-22) took place on November 23, 2025. Kevin serves as an advisor and has been tracking Quinten's progress at [Ranger](companies/ranger-22) since the early days. The meeting was scheduled as a check-in following Ranger's recent Series B announcement and to discuss some organizational challenges that have emerged during rapid scaling.\n\nQuinten opened by sharing that the engineering team has grown from 8 to 31 people in under a year, and the cultural strain is becoming noticable. Several early employees have expressed frustration with the new processes being implemented. Kevin pushed back on Quinten's instinct to slow hiring, arguing that the health tech space moves too fast to pump the brakes now. He referenced his own experience scaling teams at previous companies and suggested Quinten focus on promoting from within rather than bringing in external engineering managers.\n\nDiscussion turned to Ranger's core product roadmap. The chronic disease management platform has gained significant traction with regional hospital networks, but enterprise sales cycles remain painfully long. Quinten mentioned they're exploring a self-serve tier for smaller clinics, which Kevin thought was a distraction. \"You're not a PLG company,\" he said. \"Double down on what's working.\" Quinten seemed unconvinced but agreed to table the idea until Q2.\n\nKevin asked about the board dynamics following the new lead investor joining. Quinten admitted there's been some tension around burn rate expectations. The new board member wants to see path to profitability sooner than the original plan outlined. Kevin offered to have a sidebar conversation with the investor, having worked with them before on another deal.\n\nThe meeting wrapped with personal check-ins. Quinten mentioned he's been sleeping poorly and his partner has been frustrated with his work hours. Kevin shared some of his own experiences with founder burnout and recommended Quinten actually take the vacation he's been postponing. They agreed to meet again in December after the holiday break.",
-  "timeline": "- **2023-06-14** | Kevin Taylor first introduced to Quinten Rodriguez through mutual connection at health tech mixer\n- **2023-09-20** | Kevin agrees to advisory role at [Ranger](companies/ranger-22) following seed round close\n- **2024-02-08** | Monthly 1:1 cadence established between Kevin and Quinten\n- **2024-07-15** | Kevin helps Quinten prepare for Series A pitch, provides intro to Baseline Ventures\n- **2024-11-03** | Discussion about early employee equity refresh during scaling phase\n- **2025-04-22** | [Kevin Taylor](people/kevin-taylor-102) advises on VP Engineering search criteria\n- **2025-08-30** | Quinten seeks Kevin's input on Series B term sheet negotiations\n- **2025-10-17** | Brief call re: new board member onboarding concerns\n- **2025-11-23** | This 1:1 meeting covering scaling challenges and product strategy",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-22-2025-11-23",
-    "name": "1:1 Kevin Taylor + Quinten Rodriguez",
-    "meeting_type": "one_on_one",
-    "date": "2025-11-23",
-    "attendees": [
-      "people/kevin-taylor-102",
-      "people/quinten-rodriguez-22"
-    ],
-    "topic_company": "companies/ranger-22"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-23-2025-12-24.json b/eval/data/world-v1/meetings__oneonone-23-2025-12-24.json
deleted file mode 100644
index 9fdf5751d..000000000
--- a/eval/data/world-v1/meetings__oneonone-23-2025-12-24.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-23-2025-12-24",
-  "type": "meeting",
-  "title": "1:1 Vera Gonzalez + Paul Anderson",
-  "compiled_truth": "This one-on-one between [Vera Gonzalez](people/vera-gonzalez-103) and [Paul Anderson](people/paul-anderson-23) took place on Christmas Eve 2025, which is notable in itself—both clearly prioritizing the conversation over holiday prep. Vera had been pushing for this meeting for weeks, wanting to get Paul's perspective on the Series B dynamics at [Sentinel](companies/sentinel-23) before year-end planning kicked off.\n\nPaul came prepared with concerns about burn rate. Sentinel's consumer social play has been expensive—user acquisition costs running 40% higher than projections, and the retention numbers aren't where they need to be. He's seen this movie before at his last two startups and wanted to flag it early. Vera appreciated the candor but pushed back on some of his assumptions, arguing that the holiday season numbers would normalize by Q1.\n\nThe conversation shifted to team dynamics around the 45-minute mark. There's been some tension between the growth team and product, with each blaming the other for the conversion funnel issues. Paul thinks its a leadership gap—they need a strong VP Product who can bridge both worlds. Vera mentioned she's been talking to a few candidates but nothing concrete yet.\n\nOne interesting thread: Paul floated the idea of pivoting Sentinel toward a more niche community model rather than chasing mass-market social. He's been watching the fragmentation of social media and thinks there's more defensibility in going deep rather than wide. Vera wasn't dismissive but noted the board would need convincing, especially given the growth metrics they promised investors.\n\nThey agreed to reconvene in January with a more detailed proposal. Paul will put together a financial model showing both scenarios—continue current trajectory vs. niche pivot. Vera committed to having preliminary conversations with two board members to gauge appetite for strategic shifts. The meeting ran over by about twenty minutes, ending with some brief holiday well-wishes.",
-  "timeline": "- **2024-03-15** | Paul Anderson joins [Sentinel](companies/sentinel-23) as Head of Operations after introduction from mutual contact\n- **2024-06-22** | First informal coffee between Vera and Paul to discuss growth strategy concerns\n- **2024-09-10** | [Vera Gonzalez](people/vera-gonzalez-103) promoted to Chief of Staff, begins regular 1:1 cadence with senior leaders\n- **2025-02-28** | Paul raises burn rate concerns in leadership offsite, tension with CFO noted\n- **2025-07-14** | Joint presentation by Vera and Paul to board on operational efficiency initiatives\n- **2025-10-03** | Paul shares pivot thesis privately with Vera over dinner, asks her to keep it confidential\n- **2025-12-24** | This 1:1 meeting held on Christmas Eve to discuss Series B dynamics and potential strategic pivot",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-23-2025-12-24",
-    "name": "1:1 Vera Gonzalez + Paul Anderson",
-    "meeting_type": "one_on_one",
-    "date": "2025-12-24",
-    "attendees": [
-      "people/vera-gonzalez-103",
-      "people/paul-anderson-23"
-    ],
-    "topic_company": "companies/sentinel-23"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-24-2025-01-25.json b/eval/data/world-v1/meetings__oneonone-24-2025-01-25.json
deleted file mode 100644
index c38c42395..000000000
--- a/eval/data/world-v1/meetings__oneonone-24-2025-01-25.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-24-2025-01-25",
-  "type": "meeting",
-  "title": "1:1 Sarah Wang + Quinten Lee",
-  "compiled_truth": "One-on-one meeting between [Sarah Wang](people/sarah-wang-104) and [Quinten Lee](people/quinten-lee-24) to discuss Quinten's progress at [Tempo](companies/tempo-24) and broader career trajectory. Sarah has been mentoring Quinten for about eight months now, since he joined the biotech startup as a senior ML engineer.\n\nQuinten came prepared with a list of concerns about the current sprint priorites. He's feeling stretched thin between the protein folding optimization work and the data pipeline overhaul that leadership wants done by Q2. Sarah pushed back gently—reminded him that saying no is a skill, and that he should escalate capacity constraints to his manager rather than just absorbing them. Classic Quinten move to try and do everything himself.\n\nThey spent a good chunk of time talking about Tempo's Series B dynamics. Sarah shared some context she'd picked up from other founders about how the fundraising environment has shifted. Quinten mentioned that the CEO has been more stressed lately, which tracks. She advised him to stay focused on shipping and not get distracted by the funding noise—his job is to build, not to worry about runway.\n\nOn the career front, Quinten brought up his interest in eventually moving into a technical leadership role. Maybe eng manager, maybe staff engineer track. Sarah asked him what he actually wants—the people management piece or the technical influence piece. He wasn't sure. She suggested he find opportunities to lead a small project end-to-end this quarter to test his appetite for coordination work.\n\nMeeting wrapped with a quick check-in on Quinten's work-life balance. He admitted he's been grinding pretty hard, weekends included. Sarah reminded him that burnout compounds and encouraged him to take at least one full day off per week. They agreed to meet again in three weeks.",
-  "timeline": "- **2024-05-12** | Sarah and Quinten had their first mentorship session after connecting through mutual friend at a biotech mixer\n- **2024-07-20** | Discussed Quinten's onboarding experience at [Tempo](companies/tempo-24), first 90 days reflections\n- **2024-09-08** | [Sarah Wang](people/sarah-wang-104) helped Quinten think through a difficult conversation with his tech lead\n- **2024-11-14** | Quinten shared news about shipping Tempo's first production ML model, Sarah celebrated the win\n- **2025-01-25** | This meeting—covered capacity concerns, Series B dynamics, career pathing toward technical leadership\n- **2025-02-15** | Follow-up scheduled to check in on project leadership experiment",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-24-2025-01-25",
-    "name": "1:1 Sarah Wang + Quinten Lee",
-    "meeting_type": "one_on_one",
-    "date": "2025-01-25",
-    "attendees": [
-      "people/sarah-wang-104",
-      "people/quinten-lee-24"
-    ],
-    "topic_company": "companies/tempo-24"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-3-2025-04-04.json b/eval/data/world-v1/meetings__oneonone-3-2025-04-04.json
deleted file mode 100644
index 1c3ec41b0..000000000
--- a/eval/data/world-v1/meetings__oneonone-3-2025-04-04.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-3-2025-04-04",
-  "type": "meeting",
-  "title": "1:1 David Zhang + Victor Wilson",
-  "compiled_truth": "This one-on-one between [David Zhang](people/david-zhang-83) and [Victor Wilson](people/victor-wilson-3) took place on April 4th, 2025, primarily to discuss Delta's Series B positioning and the technical due diligence process that several potential investors have been requesting.\n\nVictor opened by walking through the latest investor conversations. Three firms are currently in active diligence, with Andreessen showing the most momentum. The sticking point remains [Delta](companies/delta-3)'s regulatory pathway—specifically whether their AI-driven protein folding platform can achieve FDA clearance for diagnostic applications within the 18-month timeline they've projected. David pushed back on some of the assumptions in the pitch deck, noting that the computational biology team is already stretched thin and adding enterprise customers before the infrastructure scales could be risky.\n\nThey spent a good chunk of time on hiring priorities. Victor wants to bring on a VP of Regulatory Affairs before the round closes, arguing it de-risks the story for investors. David is more focused on engineering bandwidth, particulary around the ML pipeline that's been causing latency issues in production. They agreed to prioritize the regulatory hire but also greenlight two senior ML engineers in Q2.\n\nConversation shifted to board dynamics. [David Zhang](people/david-zhang-83) expressed some frustration with how the last board meeting went—felt like the outside directors were too focused on short-term metrics rather than the platform play. Victor suggested bringing more detailed technical roadmaps to future meetings to reframe the narrative around long-term value creation.\n\nOther topics touched on: Delta's partnership discussions with a major pharma company (still under NDA), the need to refresh the data room before term sheets come in, and whether to attend the JP Morgan Healthcare Conference in January. Victor will handle the investor follow-ups while David focuses on stabilizing the platform before the diligence calls ramp up. Next check-in scheduled for April 18th.",
-  "timeline": "- **2024-09-15** | Initial intro call between David and Victor to discuss potential advisory role\n- **2024-11-02** | Victor officially joins [Delta](companies/delta-3) as Chief Business Officer\n- **2025-01-22** | First formal 1:1 to align on Series B strategy and timeline\n- **2025-02-10** | Joint presentation to board on fundraising approach\n- **2025-03-05** | Victor closes first diligence meeting with a]16z team\n- **2025-03-21** | David and Victor disagree on hiring priorities in planning session\n- **2025-04-04** | This meeting — discussed regulatory hire, investor updates, board dynamics\n- **2025-04-18** | Follow-up 1:1 scheduled to review term sheet progress",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-3-2025-04-04",
-    "name": "1:1 David Zhang + Victor Wilson",
-    "meeting_type": "one_on_one",
-    "date": "2025-04-04",
-    "attendees": [
-      "people/david-zhang-83",
-      "people/victor-wilson-3"
-    ],
-    "topic_company": "companies/delta-3"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-4-2025-05-05.json b/eval/data/world-v1/meetings__oneonone-4-2025-05-05.json
deleted file mode 100644
index 20998bc62..000000000
--- a/eval/data/world-v1/meetings__oneonone-4-2025-05-05.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-4-2025-05-05",
-  "type": "meeting",
-  "title": "1:1 Sarah Lopez + Paul Rodriguez",
-  "compiled_truth": "Weekly sync between [Sarah Lopez](people/sarah-lopez-84) and [Paul Rodriguez](people/paul-rodriguez-4) held on May 5th, 2025. This was a longer session than usual—ran about 50 minutes instead of the typical 30.\n\nPaul opened with an update on the [Epsilon](companies/epsilon-4) deal, which has been consuming most of his bandwidth for the past three weeks. The cybersecurity firm is moving faster than expected on their Series B evaluation, and they've narrowed vendors down to us plus one competitor (likely CrowdStrike based on Paul's intel). He's cautiously optimistic but flagged that their CISO has been harder to pin down lately. Sarah suggested looping in technical pre-sales earlier than planned—maybe get a demo environment spun up this week rather than waiting for formal approval.\n\nCompensation came up again. Paul mentioned he'd been approached by a recruiter, not fishing for a counter but wanted to be transparent about it. Sarah appreciated the candor. They agreed to revisit his comp package after Q2 closes, assuming the Epsilon deal and at least two others land. Paul seemed satisfied with that framing for now, though Sarah noted in her personal follow-ups that she needs to escalate this to leadership sooner rather than later—losing Paul mid-cycle would be rough.\n\nOn the personal side, Paul's wife is due in August. He's starting to think about paternity leave logistics. Sarah reminded him the company policy is 12 weeks fully paid, and they can stagger it however works best. He's leaning toward taking 4 weeks immediately then the rest spread across the fall.\n\nAction items coming out of this: Sarah to intro Paul to Jamie from solutions engineering re: Epsilon technical demo. Paul to send over his updated forecast by EOD Wednesday. Sarah to draft talking points for leadership about retention risks on the enterprise team.\n\nOverall good energy in the meeting despite some heavy topics. Paul remains one of the strongest performers on the team.",
-  "timeline": "- **2024-09-16** | Paul Rodriguez joined Sarah's team after internal transfer from mid-market sales\n- **2024-11-04** | First 1:1 between Sarah and Paul; set bi-weekly cadence initially\n- **2025-01-13** | Shifted to weekly 1:1s as Paul took on larger enterprise accounts\n- **2025-02-24** | Paul first mentioned [Epsilon](companies/epsilon-4) as a potential target account\n- **2025-03-17** | Initial discovery call with Epsilon completed; Paul reported strong fit\n- **2025-04-14** | Sarah flagged Paul as flight risk in leadership sync; compensation review requested\n- **2025-05-05** | This meeting—discussed Epsilon progress, comp concerns, paternity planning\n- **2025-05-12** | Scheduled follow-up to review technical demo feedback from Epsilon",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-4-2025-05-05",
-    "name": "1:1 Sarah Lopez + Paul Rodriguez",
-    "meeting_type": "one_on_one",
-    "date": "2025-05-05",
-    "attendees": [
-      "people/sarah-lopez-84",
-      "people/paul-rodriguez-4"
-    ],
-    "topic_company": "companies/epsilon-4"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-5-2025-06-06.json b/eval/data/world-v1/meetings__oneonone-5-2025-06-06.json
deleted file mode 100644
index 588c3b89c..000000000
--- a/eval/data/world-v1/meetings__oneonone-5-2025-06-06.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-5-2025-06-06",
-  "type": "meeting",
-  "title": "1:1 Priya Taylor + Mia Anderson",
-  "compiled_truth": "[Priya Taylor](people/priya-taylor-85) and [Mia Anderson](people/mia-anderson-5) have been meeting regularly since Mia joined [Nimbus](companies/nimbus-5) as a senior engineer on the data infrastructure team. This particular 1:1 focused heavily on Mia's career trajectory and some concerns she's been having about the upcoming reorg.\n\nPriya opened by asking about the carbon accounting pipeline migration that Mia's been leading. Mia reported that the migration is about 80% complete, though there have been some unexpected issues with the legacy data validation layer. She mentioned needing an additional two weeks beyond the original timeline, which Priya approved without much pushback. The climate data ingestion piece is apparently the trickiest part—third-party APIs have been unreliable and Mia's team has had to build more redundancy than anticipated.\n\nThe conversation shifted to Mia's concerns about where she fits in the new org structure. With Nimbus growing so quickly, there's been talk of splitting the data team into two groups: one focused on customer-facing analytics and another on internal climate modeling infrastructure. Mia expressed interest in the modeling side but worried she doesn't have enough visibility with the leadership team. Priya suggested she present at the next all-hands to showcase the pipeline work—good exposure opportunity.\n\nMia also brought up compensation. She's been approached by a competitor (didn't name which one) and wanted to understand her equity situation better. Priya committed to pulling together a total comp breakdown and scheduling a follow-up conversation with HR before end of month. She seemed genuinley supportive but also a bit caught off guard by the timing.\n\nThey wrapped up discussing Mia's PTO plans—she's taking two weeks in July to visit family in Portland. Priya reminded her to document the pipeline handoff proceedures before she leaves. Overall productive session, though the compensation discussion probably needs more attention soon.",
-  "timeline": "- **2024-09-15** | [Mia Anderson](people/mia-anderson-5) joined Nimbus, first 1:1 scheduled with Priya\n- **2024-11-08** | Discussed Mia's onboarding progress and initial project assignments\n- **2025-01-22** | Mid-cycle performance check-in, Mia received positive feedback on API work\n- **2025-03-10** | Carbon pipeline migration project officially kicked off with Mia as lead\n- **2025-04-18** | 1:1 focused on migration blockers and team resource constraints\n- **2025-05-09** | [Priya Taylor](people/priya-taylor-85) mentioned potential reorg during weekly sync\n- **2025-06-06** | Current 1:1 covering career concerns, compensation, and PTO planning\n- **2025-06-20** | Follow-up comp discussion scheduled with HR involvement",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-5-2025-06-06",
-    "name": "1:1 Priya Taylor + Mia Anderson",
-    "meeting_type": "one_on_one",
-    "date": "2025-06-06",
-    "attendees": [
-      "people/priya-taylor-85",
-      "people/mia-anderson-5"
-    ],
-    "topic_company": "companies/nimbus-5"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-6-2025-07-07.json b/eval/data/world-v1/meetings__oneonone-6-2025-07-07.json
deleted file mode 100644
index 53a25fce0..000000000
--- a/eval/data/world-v1/meetings__oneonone-6-2025-07-07.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-6-2025-07-07",
-  "type": "meeting",
-  "title": "1:1 Julia Davis + Uma Brown",
-  "compiled_truth": "This one-on-one between [Julia Davis](people/julia-davis-86) and [Uma Brown](people/uma-brown-6) took place on July 7th, 2025 to discuss Uma's transition into a more senior engineering role at [Vector](companies/vector-6). Julia, as VP of Engineering, wanted to check in on Uma's workload and get her perspective on the team's velocity heading into Q3.\n\nUma raised concerns about technical debt accumulating in the patient data pipeline. She's been flagging this for months apparently, and Julia acknowledged they've been kicking the can down the road. The health tech compliance requirements make refactoring risky mid-cycle, but Uma proposed a phased approach that wouldn't disrupt the current sprint. Julia seemed receptive, asked her to draft a proposal by end of week.\n\nThey also discussed Uma's career trajectory. She's been at Vector for almost two years now and is eager to take on more ownership. Julia mentioned the possibilty of Uma leading the new integration workstream with hospital partners—a high-visibility project that could set her up for staff engineer consideration by year end. Uma expressed interest but wanted clarity on scope before commiting.\n\nSome tension around the recent reorg came up briefly. Uma mentioned that a few engineers on her team felt blindsided by the reporting changes. Julia took responsibility for the communication gaps and said leadership is working on better change management processes. She asked Uma to be a bridge with the team, help smooth things over.\n\nAction items: Uma to send tech debt proposal by Friday. Julia to share integration workstream brief by Wednesday. Follow-up scheduled for July 21st to revisit both topics. Overall productive conversation—Julia noted she appreciates Uma's directness and willingness to push back constructively.",
-  "timeline": "- **2023-09-15** | Uma Brown joins [Vector](companies/vector-6) as a senior software engineer on the data platform team\n- **2024-02-20** | Julia Davis promoted to VP of Engineering at Vector\n- **2024-06-10** | First formal 1:1 between Julia and Uma after reorg placed Uma's team under Julia's org\n- **2024-11-08** | Uma presents patient data pipeline architecture at Vector all-hands\n- **2025-03-03** | [Julia Davis](people/julia-davis-86) nominates Uma for engineering excellence award\n- **2025-05-19** | Uma raises tech debt concerns in team retro, escalates to Julia\n- **2025-07-07** | This 1:1 meeting discussing career growth and technical priorities\n- **2025-07-21** | Follow-up 1:1 scheduled to review integration workstream proposal",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-6-2025-07-07",
-    "name": "1:1 Julia Davis + Uma Brown",
-    "meeting_type": "one_on_one",
-    "date": "2025-07-07",
-    "attendees": [
-      "people/julia-davis-86",
-      "people/uma-brown-6"
-    ],
-    "topic_company": "companies/vector-6"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-7-2025-08-08.json b/eval/data/world-v1/meetings__oneonone-7-2025-08-08.json
deleted file mode 100644
index c41610973..000000000
--- a/eval/data/world-v1/meetings__oneonone-7-2025-08-08.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-7-2025-08-08",
-  "type": "meeting",
-  "title": "1:1 Helen Martinez + Ulrich Johnson",
-  "compiled_truth": "One-on-one meeting between [Helen Martinez](people/helen-martinez-87) and [Ulrich Johnson](people/ulrich-johnson-7) to discuss the ongoing [Quantum](companies/quantum-7) evaluation. Helen came prepared with a detailed breakdown of their technical architecture concerns, particularly around the real-time payment processing layer that Quantum claims can handle 50k transactions per second.\n\nUlrich opened by noting that he'd spoken with two former Quantum engineers over the past week, and the picture they painted was mixed. The core technology is genuinely innovative—their approach to distributed ledger reconcilliation is unlike anything else in the fintech space—but there are questions about the leadership team's ability to scale operations. The CTO apparently has a reputation for being brilliant but difficult, which has led to significant turnover in the engineering org.\n\nHelen pushed back somewhat, arguing that founder-led technical vision often comes with rough edges and that shouldn't necessarily be a dealbreaker. She referenced the Meridian deal from 2023 where similar concerns were raised and ultimately proved overblown. Ulrich acknowledged the point but noted that Quantum's burn rate is concerningly high given their current revenue trajectory.\n\nKey discussion points centered on three areas: 1) whether the Series B terms being offered are reasonable given market conditions, 2) if the regulatory tailwinds in embedded finance are strong enough to offset execution risk, and 3) what role Helen's previous relationship with the Quantum CFO should play in the diligence process. [Ulrich Johnson](people/ulrich-johnson-7) suggested bringing in outside counsel to review the compliance framework before any term sheet is signed.\n\nThe meeting ran about fifteen minutes over as they got into the weeds on competitive positioning. Both agreed that Quantum's main vulnerability is the potential for larger players like Stripe or Plaid to build similar capabilties in-house. Next steps: Helen will schedule a follow-up call with Quantum's head of product, and Ulrich will circulate his notes from the engineer conversations to the broader investment committee.",
-  "timeline": "- **2025-07-22** | Initial intro meeting with [Quantum](companies/quantum-7) leadership team\n- **2025-07-29** | [Helen Martinez](people/helen-martinez-87) conducts preliminary technical due diligence call\n- **2025-08-01** | Ulrich begins backchannel reference checks with former Quantum employees\n- **2025-08-05** | Quantum sends over updated financials and Series B term sheet draft\n- **2025-08-08** | This 1:1 meeting to align on evaluation status and concerns\n- **2025-08-12** | Scheduled follow-up: Helen call with Quantum head of product\n- **2025-08-15** | Investment committee review tentatively planned",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-7-2025-08-08",
-    "name": "1:1 Helen Martinez + Ulrich Johnson",
-    "meeting_type": "one_on_one",
-    "date": "2025-08-08",
-    "attendees": [
-      "people/helen-martinez-87",
-      "people/ulrich-johnson-7"
-    ],
-    "topic_company": "companies/quantum-7"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-8-2025-09-09.json b/eval/data/world-v1/meetings__oneonone-8-2025-09-09.json
deleted file mode 100644
index 86d5b2f23..000000000
--- a/eval/data/world-v1/meetings__oneonone-8-2025-09-09.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-8-2025-09-09",
-  "type": "meeting",
-  "title": "1:1 Fiona Moore + Yara Johnson",
-  "compiled_truth": "This one-on-one between [Fiona Moore](people/fiona-moore-88) and [Yara Johnson](people/yara-johnson-8) took place on September 9th, 2025, as part of their regular sync cadence. Fiona, who serves as VP of Product at [Pulse](companies/pulse-8), has been meeting with Yara bi-weekly since Yara joined the company earlier this year.\n\nThe meeting opened with a check-in on Yara's onboarding progress. She's now three months in and feeling more confident navigating the codebase, though she flagged some friction with the current documentation around the adaptive learning engine. Fiona noted this down—it's the third time someone's mentioned docs being out of date, so she's planning to bring it up at the next product-eng sync.\n\nBulk of the conversation centered on the upcomming Q4 roadmap. Yara shared her perspective on prioritization, specifically advocating for more investment in the teacher dashboard analytics. She's been talking to district admins during user interviews and keeps hearing that principals want better visibility into student engagement patterns. Fiona pushed back gently, asking how this fits against the parent portal work that's already been committed. They landed on Yara drafting a one-pager comparing both initiatives by end of week.\n\nThere was also discussion about team dynamics. Yara mentioned feeling like she's still building trust with some of the senior engineers, particularly around technical decisions. Fiona offered to facilitate a working session where Yara could walk through her design rationale for the notification system refactor. \"Sometimes people just need to see how you think,\" Fiona said.\n\nToward the end, they briefly touched on career development. Yara's interested in eventually moving into a tech lead role. Fiona suggested she start by taking point on the next sprint demo and maybe mentoring the new intern starting in October. Small steps, but visible ones.\n\nMeeting wrapped after about 40 minutes. Action items: Yara to send the prioritization doc, Fiona to schedule the eng working session.",
-  "timeline": "- **2025-06-02** | [Yara Johnson](people/yara-johnson-8) joins [Pulse](companies/pulse-8) as Senior Software Engineer\n- **2025-06-16** | First 1:1 between Yara and [Fiona Moore](people/fiona-moore-88) to establish working relationship\n- **2025-07-14** | Yara completes onboarding project, ships first feature to production\n- **2025-08-11** | Fiona and Yara discuss Q3 retrospective findings and areas for growth\n- **2025-08-25** | Yara leads her first user interview session with district administrators\n- **2025-09-09** | This meeting—roadmap prioritization, team dynamics, career development discussed\n- **2025-09-12** | Yara submits one-pager comparing teacher dashboard vs parent portal initiatives\n- **2025-09-22** | Scheduled eng working session for Yara to present notification system design",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-8-2025-09-09",
-    "name": "1:1 Fiona Moore + Yara Johnson",
-    "meeting_type": "one_on_one",
-    "date": "2025-09-09",
-    "attendees": [
-      "people/fiona-moore-88",
-      "people/yara-johnson-8"
-    ],
-    "topic_company": "companies/pulse-8"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/meetings__oneonone-9-2025-10-10.json b/eval/data/world-v1/meetings__oneonone-9-2025-10-10.json
deleted file mode 100644
index af6abe83d..000000000
--- a/eval/data/world-v1/meetings__oneonone-9-2025-10-10.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "meetings/oneonone-9-2025-10-10",
-  "type": "meeting",
-  "title": "1:1 Jack Davis + Rachel Garcia",
-  "compiled_truth": "Weekly one-on-one between [Jack Davis](people/jack-davis-89), VP of Engineering at [Helix](companies/helix-9), and Rachel Garcia, Senior Staff Engineer leading the distributed systems team. Meeting held October 10th, 2025 at Helix's SF headquarters.\n\nRachel came prepared with concerns about the upcoming v3.0 release timeline. The team has been pushing hard on the new inference routing layer, but she's seeing burnout signs across her direct reports. Jack acknowledged this and suggested they might need to have an honest conversation with leadership about scope. He mentioned that the board presentation went well last week, but there's pressure to ship before re:Invent in December.\n\nDiscussion moved to the promotion packet for Marcus Chen. Rachel believes he's ready for Staff level—his work on the caching subsystem saved them nearly $2M in compute costs last quarter. Jack agreed to champion the packet in the next calibration cycle but warned that headcount constraints might make it competitive. They talked through how to position Marcus's impact relative to other candidates.\n\nThere was some friction around the new oncall rotation. Rachel pushed back on having her team cover weekends for the platform team while they're shortstaffed. Jack said he'd escalate to get contractors approved, though he wasn't optimistic given the recent budget review. They compromised on a temporary two-week coverage arrangement.\n\n[Jack Davis](people/jack-davis-89) also brought up the technical debt initiative. He wants Rachel to lead a working group that would audit the entire inference pipeline and propose a 6-month cleanup roadmap. She's interested but concerned about bandwidth. They agreed to revisit after the v3.0 ship.\n\nAction items: Jack to follow up with HR on the contractor req, Rachel to draft the promotion packet by end of next week, both to sync with product on scope discussions Monday.",
-  "timeline": "- **2025-08-15** | Jack and Rachel started doing bi-weekly 1:1s after her promotion to Senior Staff\n- **2025-09-05** | First discussion about v3.0 timeline concerns raised by Rachel's team\n- **2025-09-19** | Rachel presented distributed systems roadmap to [Jack Davis](people/jack-davis-89) and CTO\n- **2025-10-03** | Previous 1:1 focused on oncall rotation issues and team morale\n- **2025-10-10** | Current meeting covering promotion packets, release scope, and technical debt initiative\n- **2025-10-17** | Scheduled follow-up to review Marcus Chen promotion packet draft\n- **2025-11-01** | Planned scope discussion with product leadership at [Helix](companies/helix-9)",
-  "_facts": {
-    "type": "meeting",
-    "slug": "meetings/oneonone-9-2025-10-10",
-    "name": "1:1 Jack Davis + Rachel Garcia",
-    "meeting_type": "one_on_one",
-    "date": "2025-10-10",
-    "attendees": [
-      "people/jack-davis-89",
-      "people/rachel-garcia-9"
-    ],
-    "topic_company": "companies/helix-9"
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__adam-lee-19.json b/eval/data/world-v1/people__adam-lee-19.json
deleted file mode 100644
index 691087312..000000000
--- a/eval/data/world-v1/people__adam-lee-19.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/adam-lee-19",
-  "type": "person",
-  "title": "Adam Lee",
-  "compiled_truth": "Adam Lee is the founder of [Forge](companies/forge-19), a crypto infrastructure company focused on building tooling for decentralized finance protocols. He started the company in late 2021 after spending several years in traditional finance and a brief stint at a major blockchain startup. Known around the industry as a long-term thinker, Adam has repeatedly emphasized that Forge isn't chasing short-term token pumps but rather building foundational rails that will matter in ten years.\n\nAdam's leadership style is often described as demanding. Former employees have noted his tendency to push teams hard, expecting exceptional output and deep technical rigor. This has created a culture at Forge thats either loved or hated—people who thrive there tend to be self-directed engineers who appreciate direct feedback. Those who don't fit the mold typically leave within six months. Lee himself has acknowledged this dynamic in interviews, saying he'd rather have a small team of exceptional builders than a bloated org of mediocre talent.\n\nBefore [Forge](companies/forge-19), Adam worked at Goldman Sachs in their digital assets exploratory group, though he's been vocal about his frustrations with how slow traditional institutions move. He left in 2020 to join a Layer 1 project as head of protocol development, but departed after disagreements with the founding team over tokenomics decisions he viewed as short-sighted. This experience shaped his approach with Forge—he maintains majority control and has been selective about investors, preferring those aligned with his multi-year vision.\n\nLee holds a CS degree from MIT and briefly pursued a PhD before dropping out to enter industry. He's published several technical papers on cryptographic primatives and occasionally speaks at conferences, though he tends to avoid the spotlight compared to other crypto founders. His public presence is mostly limited to occasional Twitter threads and a quarterly investor letter that sometimes gets leaked to crypto media. People who know Adam describe him as intense but fair, someone who genuinely cares about the technology rather than personal enrichment.",
-  "timeline": "- **2021-11-15** | Adam Lee officially incorporates Forge, announcing the project via a detailed technical blog post\n- **2022-03-08** | Closes $12M seed round led by Paradigm; Lee insists on founder-friendly terms with minimal board seats\n- **2022-09-22** | Keynote at Mainnet conference in NYC, outlining Forge's 10-year infrastructure roadmap\n- **2023-02-14** | Forge ships v1 of their cross-chain messaging protocol to mainnet\n- **2023-07-30** | Controversial Twitter thread criticizing VC-backed token launches goes viral\n- **2024-01-11** | Meets with [Forge](companies/forge-19) team in Denver for annual strategy offsite\n- **2024-06-19** | Series A closes at $45M valuation; Adam retains 58% ownership\n- **2024-11-03** | Publishes whitepaper on MEV-resistant transaction ordering\n- **2025-04-27** | Featured in Fortune's \"40 Under 40\" crypto edition\n- **2025-09-15** | Announces Forge enterprise partnerships with two major exchanges",
-  "_facts": {
-    "type": "person",
-    "slug": "people/adam-lee-19",
-    "name": "Adam Lee",
-    "role": "founder",
-    "primary_affiliation": "companies/forge-19",
-    "notable_traits": [
-      "long-term thinker",
-      "demanding"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__adam-lopez-113.json b/eval/data/world-v1/people__adam-lopez-113.json
deleted file mode 100644
index f0324ee7f..000000000
--- a/eval/data/world-v1/people__adam-lopez-113.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "people/adam-lopez-113",
-  "type": "person",
-  "title": "Adam Lopez",
-  "compiled_truth": "Adam Lopez is a senior engineer at [Delta](companies/delta-3), a biotech company focused on synthetic biology platforms for drug discovery. He joined Delta in early 2022 after a stint at a smaller computational biology startup that got acquihired by Roche. Known internally as someone who ships fast and asks questions later, Adam has built a reputation for being deeply opinionated about technical architecture—sometimes to a fault.\n\nAdam's background is primarily in backend systems and data pipelines, though he's picked up enough wet lab knowledge to be dangerous. Colleagues describe him as the kind of engineer who'll rewrite a critical service over a weekend and have it running in production by Monday, which is either inspiring or terrifying depending on who you ask. His work on Delta's core sequencing analysis pipeline cut processing time by nearly 40%, a fact he's not shy about mentioning.\n\nBefore [Delta](companies/delta-3), Lopez worked at a YC-backed startup building developer tools for bioinformatics. That company never quite found product-market fit, but Adam learned to move fast in ambiguous enviroments—a skill that translated well to the chaotic early days at Delta. He's also done contract work for a few pharma companies, though he doesn't talk much about that period.\n\nOn the opinonated front: Adam has strong views on everything from database choices (Postgres, always) to meeting culture (should be abolished, mostly). He's written several internal memos that have become semi-legendary within Delta's engineering org, including one titled \"Why We Should Delete Half Our Microservices\" that apparently caused quite a stir. Despite the friction this sometimes creates, leadership values his willingness to challenge assumptions.\n\nOutside work, Adam is into long-distance running and occasionally posts race results that make his coworkers feel inadequate. He's based in the Bay Area but has talked about relocating to somewhere with actual seasons.",
-  "timeline": "- **2021-03-15** | Left bioinformatics startup after acquisition by Roche; took a few months off\n- **2022-01-10** | Joined [Delta](companies/delta-3) as senior backend engineer\n- **2022-08-22** | Shipped v2 of Delta's sequencing analysis pipeline; 40% speed improvement\n- **2023-02-14** | Wrote controversial internal memo on microservices architecture\n- **2023-06-30** | Promoted to staff engineer at Delta\n- **2023-11-08** | Gave internal talk on \"Moving Fast Without Breaking Biotech\"\n- **2024-04-19** | Led infrastructure migration to new compute cluster\n- **2024-09-12** | Mentored two new hires on pipeline development practices\n- **2025-01-27** | Started working on Delta's next-gen protein modeling integration",
-  "_facts": {
-    "type": "person",
-    "slug": "people/adam-lopez-113",
-    "name": "Adam Lopez",
-    "role": "engineer",
-    "primary_affiliation": "companies/delta-3",
-    "secondary_affiliations": [],
-    "notable_traits": [
-      "opinionated",
-      "fast-shipping"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__alice-davis-172.json b/eval/data/world-v1/people__alice-davis-172.json
deleted file mode 100644
index 594d72016..000000000
--- a/eval/data/world-v1/people__alice-davis-172.json
+++ /dev/null
@@ -1,21 +0,0 @@
-{
-  "slug": "people/alice-davis-172",
-  "type": "person",
-  "title": "Alice Davis",
-  "compiled_truth": "Alice Davis is an advisor at [Prism](companies/prism-43), a cybersecurity company where she's helped shape their enterprise security architecture from the ground up. Known industrywide as a systems builder, she has a reputation for constructing frameworks that actually scale—not just on paper, but in the messy reality of production environments.\n\nBefore taking on advisory roles, Alice spent over a decade in operational security leadership. She ran infrastructure teams at two mid-stage startups, both of which exited successfully. Colleagues describe her as demanding, sometimes exhaustingly so, but the results speak for themselves. Teams she's mentored have gone on to lead security orgs at major tech companies. She doesn't suffer fools, and she doesn't accept \"good enough\" when better is achievable.\n\nHer work with Prism began in late 2022 when the company was struggling to articulate their technical differentiation. Davis came in, audited their entire stack over six weeks, and produced a roadmap that the engineering team still references today. She pushed hard for a zero-trust architecture overhaul that initially met resistance from leadership, but her persistence paid off—Prism's enterprise contracts doubled within eighteen months.\n\nAlice also maintains an affiliation with [Orbit Labs](companies/orbit-labs-92), where she advises on security posture for their distributed compute platform. Her involvement there is more sporadic, typically ramping up before major releases or compliance audits. She's helped them navigate SOC 2 certification twice now.\n\nOutside of her advisory work, Davis occasionally speaks at security conferences, though she's selective about which invitations she accepts. She prefers smaller venues where she can engage directly with practitioners rather than deliver keynotes to passive audiences. Her talks tend to be practical, focused on failure modes and recovery patterns rather than abstract theory.\n\nShe lives in the Boston area, maintains a deliberately minimal social media presence, and is rumoured to be working on a book about organizational security culture—though she's never confirmed this publicly.",
-  "timeline": "- **2021-03-15** | Joined [Orbit Labs](companies/orbit-labs-92) as security advisor ahead of their Series B\n- **2021-09-08** | Delivered workshop on incident response frameworks at SecureCon Northeast\n- **2022-04-22** | Completed SOC 2 audit preparation for Orbit Labs\n- **2022-11-01** | Began formal advisory engagement with [Prism](companies/prism-43)\n- **2023-02-14** | Presented zero-trust architecture proposal to Prism leadership team\n- **2023-07-30** | Prism enterprise security overhaul reached production deployment\n- **2024-01-18** | Led security review for Orbit Labs v3.0 platform release\n- **2024-06-05** | Featured in CyberDefense Weekly profile on influential advisors\n- **2025-03-22** | Renewed multi-year advisory contract with Prism\n- **2025-11-10** | Spoke at private dinner event on building security-first engineering cultures",
-  "_facts": {
-    "type": "person",
-    "slug": "people/alice-davis-172",
-    "name": "Alice Davis",
-    "role": "advisor",
-    "primary_affiliation": "companies/prism-43",
-    "secondary_affiliations": [
-      "companies/orbit-labs-92"
-    ],
-    "notable_traits": [
-      "systems builder",
-      "demanding"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__beth-williams-177.json b/eval/data/world-v1/people__beth-williams-177.json
deleted file mode 100644
index 0f8e23413..000000000
--- a/eval/data/world-v1/people__beth-williams-177.json
+++ /dev/null
@@ -1,21 +0,0 @@
-{
-  "slug": "people/beth-williams-177",
-  "type": "person",
-  "title": "Beth Williams",
-  "compiled_truth": "Beth Williams is an advisor at [Echo Labs](companies/echo-labs-82), a biotech company where she's developed a reputation for being demanding and opinionated—qualities that have made her both invaluable and occasionally difficult to work with. She doesn't sugarcoat feedback, and founders who seek her counsel know they're signing up for unvarnished truth.\n\nBeth's background is in molecular biology, with a PhD from Stanford and nearly a decade in big pharma before she pivoted to the startup world. She spent five years as VP of Research at a mid-sized diagnostics company before transitioning fully into advisory roles. Her expertise lies in regulatory strategy and clinical trial design, areas where her exacting standards actually serve founders well.\n\nAt Echo Labs, Williams has been instrumental in shaping their approach to FDA engagement. The company's leadership credits her with pushing them to pursue a more agressive timeline for their lead therapeutic candidate, though not without significant internal debate. Beth is known for showing up to board meetings with marked-up documents and pointed questions that can derail agendas but often surface critical blind spots.\n\nShe also maintains an advisory relationship with [Ranger](companies/ranger-22), though her involvement there is less intensive. Sources suggest she was brought on primarily for her network of clinical investigators, which she's cultivated over two decades in the industry.\n\nColleagues describe Beth as someone who \"doesn't do small talk.\" She's focused, sometimes to a fault, and has little patience for founders who haven't done their homework. One Echo Labs exec noted that \"Beth will tell you your idea is stupid, but she'll also stay late helping you figure out how to make it less stupid.\" That combination of bluntness and genuine investment in outcomes has made her a sought-after advisor despite her rough edges.\n\nOutside of her formal roles, Williams occasionally speaks at biotech conferences, though she's selective about which invitations she accepts. She lives in the Bay Area and is reportedly an avid trail runner.",
-  "timeline": "- **2021-03-15** | Beth Williams joins [Echo Labs](companies/echo-labs-82) as a formal advisor, brought on to guide regulatory strategy\n- **2021-09-22** | Delivers keynote on clinical trial efficiency at BioConnect Summit in Boston\n- **2022-02-08** | Facilitates introduction between Echo Labs and key opinion leaders in oncology space\n- **2022-07-11** | Signs advisory agreement with [Ranger](companies/ranger-22), focusing on clinical network development\n- **2023-01-19** | Leads contentious board discussion at Echo Labs re: timeline acceleration for Phase II trials\n- **2023-08-30** | Referenced in industry profile as one of \"20 Advisors Actually Worth Listening To\"\n- **2024-04-14** | Helps Echo Labs navigate FDA pre-submission meeting prep\n- **2024-11-02** | Participates in Ranger strategic offsite, pushes back on proposed pivit to consumer health\n- **2025-06-18** | Renews advisory engagement with Echo Labs for another two-year term",
-  "_facts": {
-    "type": "person",
-    "slug": "people/beth-williams-177",
-    "name": "Beth Williams",
-    "role": "advisor",
-    "primary_affiliation": "companies/echo-labs-82",
-    "secondary_affiliations": [
-      "companies/ranger-22"
-    ],
-    "notable_traits": [
-      "demanding",
-      "opinionated"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__carol-jackson-81.json b/eval/data/world-v1/people__carol-jackson-81.json
deleted file mode 100644
index 985e3f048..000000000
--- a/eval/data/world-v1/people__carol-jackson-81.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "people/carol-jackson-81",
-  "type": "person",
-  "title": "Carol Jackson",
-  "compiled_truth": "Carol Jackson is a partner at [Sequoia Capital](companies/sequoia-capital-1), where she's built a reputation as one of the firm's most effective talent scouts and go-to-market strategists. Before joining Sequoia in 2019, Carol spent nearly a decade in operating roles—first at Stripe leading enterprise sales, then as VP of Revenue at a Series B startup that ultimately sold to Salesforce. That operational DNA shows up in how she works with founders today.\n\nHer portfolio leans heavily toward infrastructure and developer tools, though she's shown a willingness to back consumer when the GTM motion is sufficiently differentiated. Current board seats include [Prism](companies/prism-43), [Beacon Labs](companies/beacon-labs-60), and [Foundry Labs](companies/foundry-labs-83). She also serves as an observer at [Keel](companies/keel-38) and [Pulse Labs](companies/pulse-labs-58), both earlier-stage bets where she's been instrumental in shaping early hiring plans.\n\nCarol is known for rolling up her sleeves on recruiting. Founders describe her as \"relentless\" when it comes to closing executive candidates—she'll fly cross-country for a dinner if it means landing a critical VP of Engineering hire. Her network in the sales and marketing talent pool is particularly deep, a byproduct of her years building revenue teams. Multiple portfolio CEOs have credited her with sourcing their first head of sales.\n\nOn the GTM side, Jackson brings a frameworks-heavy approach. She's developed internal playbooks at Sequoia around pricing strategy, sales compensation, and enterprise readiness assessments. These get deployed across her portfolio companies, often during the Series A to B transition when go-to-market complexity spikes. Some founders find the structure invaluable; others have noted it can feel prescriptive.\n\nOutside of investing, Carol is a frequent speaker at SaaStr and other industry events. She's written extensively about the \"sales-led to product-led\" transition, arguing that most companies botch it by moving too early. Her Twitter presence is modest but engaged—she tends to amplify founder content rather than posting hot takes.\n\nShe lives in Menlo Park with her husband and two kids. Known to be an early riser, often sending emails before 6am PT.",
-  "timeline": "- **2021-03-15** | Led Series A for [Prism](companies/prism-43), her first solo deal at Sequoia after co-leading several investments\n- **2021-09-22** | Keynote at SaaStr Annual on \"Why Your First Sales Hire Should Be a Builder, Not a Closer\"\n- **2022-02-10** | Joined board of [Beacon Labs](companies/beacon-labs-60) following their $28M Series B\n- **2022-08-04** | Helped recruit VP of Sales at [Keel](companies/keel-38), closing candidate after 3-month search\n- **2023-01-19** | Published internal memo on AI's impact on GTM roles, later leaked and discussed widely on Twitter\n- **2023-06-30** | Led preemptive Series B for [Foundry Labs](companies/foundry-labs-83) at $180M valuation\n- **2024-02-14** | Took observer seat at [Pulse Labs](companies/pulse-labs-58) as part of seed extension\n- **2024-11-08** | Named to Forbes Midas List for first time, ranked #47\n- **2025-04-22** | Hosted private dinner with 12 portfolio CEOs in SF to discuss tariff impacts on hardware startups",
-  "_facts": {
-    "type": "person",
-    "slug": "people/carol-jackson-81",
-    "name": "Carol Jackson",
-    "role": "partner",
-    "primary_affiliation": "companies/sequoia-capital-1",
-    "secondary_affiliations": [
-      "companies/prism-43",
-      "companies/beacon-labs-60",
-      "companies/foundry-labs-83",
-      "companies/keel-38",
-      "companies/pulse-labs-58"
-    ],
-    "notable_traits": [
-      "recruiting strength",
-      "GTM-heavy"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__carol-wilson-28.json b/eval/data/world-v1/people__carol-wilson-28.json
deleted file mode 100644
index 84b8371e6..000000000
--- a/eval/data/world-v1/people__carol-wilson-28.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/carol-wilson-28",
-  "type": "person",
-  "title": "Carol Wilson",
-  "compiled_truth": "Carol Wilson is the founder of [Anchor](companies/anchor-28), a data infrastructure company that's been quietly building out tooling for enterprise data pipelines. She's known in founder circles for her unusual patience—willing to spend years on go-to-market rather than rushing to scale prematurely. This distribution-focused mindset has shaped Anchor's entire strategy.\n\nBefore starting Anchor, Carol spent nearly a decade at Snowflake, where she led partnerships and saw firsthand how critical distribution channels were for infrastructure products. She left in 2021, convinced that the next wave of data tooling would need fundamentally different GTM motions than the previous generation. Her thesis: most data infrastructure companies fail not because of product, but because they don't understand how to reach the right buyers at the right time.\n\nWilson is notably private for a founder. She rarely speaks at conferences, preferring smaller roundtables and direct conversations with potential customers. When she does appear publicly, she tends to focus on the unsexy parts of building—pricing strategy, sales team composition, channel partnerships. Colleagues describe her as methodical, sometimes frustratingly so. One early employee noted that Carol would rather delay a launch by six months than ship to the wrong segment.\n\nAnchor itself reflects this philosophy. The company spent its first eighteen months in near-stealth, working with just a handful of design partners before opening up more broadly. Their core product handles data pipeline orchestration, but the real differentiator according to Wilson is their integration approach—they've built connectors to dozens of enterprise systems, making adoption friction minimal.\n\nCarol holds a CS degree from University of Washington and an MBA from Stanford. She's based in Seattle and has been an angel investor in about a dozen companies, mostly in the data and developer tools space. Her investments tend toward technical founders who she thinks underestimate distribution—she sees herself as a resource for helping them avoid common mistakes. She sits on no outside boards currently, keeping her focus squarely on [Anchor](companies/anchor-28) as it enters what she calls the \"real scaling phase.\"",
-  "timeline": "- **2021-03-15** | Carol Wilson leaves Snowflake after 9 years to begin working on stealth data infrastructure concept\n- **2021-08-02** | Incorporates [Anchor](companies/anchor-28) in Delaware, begins recruiting founding team\n- **2022-01-19** | Anchor closes $4.2M seed round, Carol opts for smaller raise to maintain patience on GTM\n- **2022-09-11** | First design partner signed—a Fortune 500 retailer needing pipeline orchestration\n- **2023-04-28** | Wilson speaks at private data infrastructure roundtable in SF, emphasizes distribution over product velocity\n- **2023-11-06** | Anchor quietly opens product to broader waitlist after 18 months in controlled access\n- **2024-02-14** | Series A closes at $18M, Carol negotiates unusually founder-friendly terms\n- **2024-08-30** | Carol begins angel investing more actively, makes three investments in Q3\n- **2025-01-22** | Anchor crosses 100 enterprise customers, Wilson notes they're \"just getting started\"\n- **2025-05-17** | Featured in The Information piece on patient founders defying growth-at-all-costs mentality",
-  "_facts": {
-    "type": "person",
-    "slug": "people/carol-wilson-28",
-    "name": "Carol Wilson",
-    "role": "founder",
-    "primary_affiliation": "companies/anchor-28",
-    "notable_traits": [
-      "distribution-focused",
-      "patient"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__chris-jackson-91.json b/eval/data/world-v1/people__chris-jackson-91.json
deleted file mode 100644
index 134ff25e4..000000000
--- a/eval/data/world-v1/people__chris-jackson-91.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "people/chris-jackson-91",
-  "type": "person",
-  "title": "Chris Jackson",
-  "compiled_truth": "Chris Jackson is a partner at [Initialized](companies/initialized-11), where he's built a reputation as one of the more opinionated voices in early-stage venture. He doesn't hedge — if he thinks a founder is chasing the wrong market, he'll say it directly. This has earned him both admirers and detractors, though most agree his analytical rigor is genuinely useful when applied constructively.\n\nBefore joining Initialized, Chris spent several years as a product lead at a mid-stage fintech that never quite broke out. That experience — watching a promising company stall due to poor capital allocation and founder indecision — shaped his investing philosophy. He's particularly focused on capital efficiency and tends to push back hard on founders who want to raise large rounds before proving unit economics. Some find this annoying. Others credit him with saving their companies from dilution death spirals.\n\nHis portfolio includes board seats at [Meridian](companies/meridian-40) and [Gravity](companies/gravity-17), both of which reflect his interest in infrastructure plays with defensible moats. He led Initialized's investment in Meridian during their seed round and has been deeply involved in their go-to-market strategy since. At Gravity, he came in later but quickly became the board member founders call when they need to think through pricing or competitive positioning.\n\nChris also has smaller advisory roles with [Acme](companies/acme-0) and [Lucid Labs](companies/lucid-labs-71), though he's less hands-on with those. He tends to reserve his energy for companies where he has real conviction.\n\nOutside of work, Jackson is known for his detailed investment memos that occasionally leak and circulate among founders. He doesn't seem to mind — if anything, he enjoys the debate they spark. He's been on a few podcasts but generally avoids the conference circuit, prefering to spend time with portfolio companies rather than on panels. Lives in SF, runs most mornings, and is apparently a decent cook, though that's secondhand information at best.",
-  "timeline": "- **2021-03-15** | Chris Jackson joins [Initialized](companies/initialized-11) as a partner after brief stint as EIR\n- **2021-09-22** | Leads seed investment in [Meridian](companies/meridian-40), takes board seat\n- **2022-04-10** | Internal memo on \"Why Most Seed Rounds Are Too Large\" circulates widely on Twitter\n- **2022-11-03** | Joins [Gravity](companies/gravity-17) board following their Series A\n- **2023-02-18** | Speaks at small founder dinner on capital efficiency — recording gets shared around\n- **2023-08-07** | Begins informal advisory work with [Lucid Labs](companies/lucid-labs-71)\n- **2024-01-29** | Helps [Acme](companies/acme-0) restructure pricing model during crunch period\n- **2024-06-14** | Quoted in The Information piece on seed market dynamics\n- **2025-03-02** | [Meridian](companies/meridian-40) closes Series B; Chris leads board discussion on expansion timing\n- **2025-11-20** | Named to Forbes Midas List honorable mention for first time",
-  "_facts": {
-    "type": "person",
-    "slug": "people/chris-jackson-91",
-    "name": "Chris Jackson",
-    "role": "partner",
-    "primary_affiliation": "companies/initialized-11",
-    "secondary_affiliations": [
-      "companies/meridian-40",
-      "companies/gravity-17",
-      "companies/acme-0",
-      "companies/lucid-labs-71"
-    ],
-    "notable_traits": [
-      "opinionated",
-      "analytical"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__chris-singh-96.json b/eval/data/world-v1/people__chris-singh-96.json
deleted file mode 100644
index 4010980bc..000000000
--- a/eval/data/world-v1/people__chris-singh-96.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "people/chris-singh-96",
-  "type": "person",
-  "title": "Chris Singh",
-  "compiled_truth": "Chris Singh is a partner at [Sequoia Capital II](companies/sequoia-capital-ii-16), where he's built a reputation for having unusually deep technical chops combined with genuine design taste—a rare combo in the venture world. Before joining Sequoia, Chris spent nearly a decade as an engineer and later engineering lead at several startups, which gives him credibility when sitting across the table from technical founders.\n\nSingh's portfolio reflects his interests: developer tools, infrastructure, and products where craft matters. He led Sequoia's investment in [Gravity Labs](companies/gravity-labs-67) and sits on their board, often getting into the weeds with their engineering team on architecture decisions. He's also involved with [Beta Labs](companies/beta-labs-51), [Meridian Labs](companies/meridian-labs-90), and [Jolt Labs](companies/jolt-labs-87)—a mix of early and growth-stage bets across infra and consumer.\n\nChris is known for being direct, sometimes to a fault. Founders describe him as the partner who will actualy tell you when something isn't working, rather than ghosting or giving vague feedback. He runs a tight process during diligence, often asking to see codebases and Figma files before commiting to a term sheet. Some find this intrusive; others appreciate the signal that he'll be a hands-on board member.\n\nHe grew up in Toronto, studied computer science at Waterloo, and did a brief stint at Google before catching the startup bug. His first company—a mobile analytics tool—was acqui-hired by a larger adtech player in 2014. Not a home run, but enough to give him operator cred and a small financial cushion to start angel investing.\n\nOutside of work, Singh is a design nerd. He collects vintage Braun products and has strong opinions about typography. He's been spotted at Figma Config and Apple WWDC more than once, usually taking notes in a Field Notes notebook. His Twitter presence is sparse but occasionally reveals contrarian takes on product development and hiring.",
-  "timeline": "- **2021-03-15** | Chris joined [Sequoia Capital II](companies/sequoia-capital-ii-16) as a partner, transitioning from his operating roles\n- **2021-09-02** | Led Series A investment in [Gravity Labs](companies/gravity-labs-67), taking a board seat\n- **2022-04-18** | Participated in seed round for [Beta Labs](companies/beta-labs-51) alongside two other funds\n- **2022-11-07** | Gave talk at a private founders dinner on \"Why most dev tools fail at design\"\n- **2023-02-22** | Joined board observer role at [Meridian Labs](companies/meridian-labs-90) following their Series B\n- **2023-08-14** | Published internal memo on AI infrastructure investment thesis, widely circulated\n- **2024-01-09** | Led pre-seed check into [Jolt Labs](companies/jolt-labs-87), his smallest bet to date\n- **2024-06-30** | Helped recruit new CTO for Gravity Labs after previous one departed\n- **2025-03-11** | Spoke at Sequoia's annual LP meeting about portfolio construction in uncertain markets\n- **2025-10-05** | Promoted to senior partner at Sequoia Capital II",
-  "_facts": {
-    "type": "person",
-    "slug": "people/chris-singh-96",
-    "name": "Chris Singh",
-    "role": "partner",
-    "primary_affiliation": "companies/sequoia-capital-ii-16",
-    "secondary_affiliations": [
-      "companies/gravity-labs-67",
-      "companies/beta-labs-51",
-      "companies/meridian-labs-90",
-      "companies/jolt-labs-87"
-    ],
-    "notable_traits": [
-      "technical depth",
-      "design taste"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__chris-smith-110.json b/eval/data/world-v1/people__chris-smith-110.json
deleted file mode 100644
index 95de18dca..000000000
--- a/eval/data/world-v1/people__chris-smith-110.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "people/chris-smith-110",
-  "type": "person",
-  "title": "Chris Smith",
-  "compiled_truth": "Chris Smith is a senior engineer at [Acme](companies/acme-0), where he's been instrumental in building out their core robotics stack since 2022. Known internally as a sharp pattern matcher, Chris has a reputation for spotting architectural flaws before they become expensive problems. His demanding nature rubs some people the wrong way, but those who've worked closely with him generally come around—he pushes hard because he cares about shipping things that actually work.\n\nBefore joining Acme, Smith spent four years at a mid-sized automation startup that eventually got acqui-hired by Amazon. The experience left him somewhat jaded about big tech acquisitions, which partly explains why he chose to join a smaller robotics outfit. He's mentioned in a few interviews that he wanted to be somewhere he could \"touch the metal\" again, rather than just pushing configs through deployment pipelines.\n\nChris is particularly skilled at debugging complex sensor fusion issues. When [Acme](companies/acme-0) hit a wall with their perception system in late 2023, he was the one who traced the problem back to a subtle timing bug in how IMU data was being interpolated. Took him about three days of obsessive log analysis. His collegues sometimes joke that he can \"smell\" race conditions.\n\nOn the personal side, Smith keeps a relatively low profile. He's based in the Bay Area, lives somewhere in Oakland, and occasionally posts about hiking on social media. Not married as far as anyone knows. He did a brief stint giving talks at robotics meetups in 2023 but seems to have pulled back from that—possibly just too busy with Acme's push toward their next product milestone.\n\nOne thing worth noting: Chris can be blunt to the point of abrasiveness in technical discussions. New hires at Acme are sometimes warned about this. It's not personal, he just doesn't have much patience for hand-wavy explanations or premature optimization debates.",
-  "timeline": "- **2021-03-15** | Left previous role at automation startup after Amazon acquisition finalized\n- **2022-01-10** | Joined [Acme](companies/acme-0) as senior engineer on perception team\n- **2022-08-22** | Led emergency fix for sensor calibration issue affecting prototype units\n- **2023-04-07** | Gave talk at Bay Area Robotics Meetup on real-time sensor fusion\n- **2023-11-30** | Identified and resolved critical IMU timing bug that had blocked release for weeks\n- **2024-02-14** | Promoted to tech lead for [Acme](companies/acme-0) hardware integration\n- **2024-09-03** | Internal presentation on reliability lessons from field testing\n- **2025-01-18** | Started mentoring two new hires on the robotics team\n- **2025-06-11** | Attended robotics summit in Boston, met with potential sensor vendors",
-  "_facts": {
-    "type": "person",
-    "slug": "people/chris-smith-110",
-    "name": "Chris Smith",
-    "role": "engineer",
-    "primary_affiliation": "companies/acme-0",
-    "secondary_affiliations": [],
-    "notable_traits": [
-      "sharp pattern matcher",
-      "demanding"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__chris-williams-37.json b/eval/data/world-v1/people__chris-williams-37.json
deleted file mode 100644
index 0cfee8488..000000000
--- a/eval/data/world-v1/people__chris-williams-37.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/chris-williams-37",
-  "type": "person",
-  "title": "Chris Williams",
-  "compiled_truth": "Chris Williams is the founder of [Jolt](companies/jolt-37), an AI applications company building tools that help enterprises automate complex decision-making workflows. Before starting Jolt, Chris spent nearly a decade in product roles at mid-stage startups, most notably leading platform strategy at a fintech company that eventually sold to a major bank. He's known among peers as a first-principles thinker—someone who refuses to accept conventional wisdom without interrogating the underlying assumptions.\n\nChris launched [Jolt](companies/jolt-37) in late 2022 after becoming convinced that most enterprise AI tooling was solving the wrong problems. Rather than chasing the hype around generative models, he focused on deterministic AI systems that could integrate with existing business logic. This patient approach meant Jolt was slower to market than competitors, but the product resonated with risk-averse industries like insurance and logistics.\n\nHe's described by collaborators as deliberate, sometimes frustratingly so. Williams will sit with a problem for weeks before committing to a direction, a habit that has occasionally caused tension with investors pushing for faster iteration. But his track record suggests the patience pays off—Jolt's retention numbers are unusually strong for a company at its stage.\n\nOutside of work, Chris is relatively private. He occasionally speaks at industry events but avoids the conference circuit. His writing, mostly on Substack, tends toward dense explorations of systems thinking and organizational design. He's mentioned in interviews that he reads more philosophy than business books, citing Wittgenstein and Simone Weil as influences on how he approaches product developement.\n\nWilliams holds a degree in cognitive science from UC San Diego. He briefly considered academia before deciding the feedback loops were too slow. Now based in Austin, he splits his time between hands-on product work and recruiting—Jolt remains a small team of around fifteen, and Chris personally interviews every hire.",
-  "timeline": "- **2022-09-14** | Chris Williams incorporates Jolt and begins building initial prototype with two cofounders\n- **2023-02-08** | Closes $2.4M pre-seed round, mostly from angel investors in the Austin tech scene\n- **2023-06-21** | Jolt ships private beta to three insurance companies for workflow automation testing\n- **2023-11-03** | Chris gives talk at AI Summit Austin on \"Why Most AI Products Fail at the Integration Layer\"\n- **2024-03-17** | [Jolt](companies/jolt-37) announces $11M seed round led by Gradient Ventures\n- **2024-07-29** | Williams publishes widely-shared essay on deterministic vs probabilistic AI systems\n- **2024-10-12** | Jolt reaches 30 enterprise customers, mostly in logistics and insurance verticals\n- **2025-01-24** | Chris begins advising two early-stage AI startups on product stratgey\n- **2025-05-09** | Featured in Forbes piece on \"patient founders\" building outside the SF bubble",
-  "_facts": {
-    "type": "person",
-    "slug": "people/chris-williams-37",
-    "name": "Chris Williams",
-    "role": "founder",
-    "primary_affiliation": "companies/jolt-37",
-    "notable_traits": [
-      "first-principles thinker",
-      "patient"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__david-wang-10.json b/eval/data/world-v1/people__david-wang-10.json
deleted file mode 100644
index e7a1bac1e..000000000
--- a/eval/data/world-v1/people__david-wang-10.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/david-wang-10",
-  "type": "person",
-  "title": "David Wang",
-  "compiled_truth": "David Wang is the founder and CEO of [Beacon](companies/beacon-10), a cybersecurity startup focused on threat detection and incident response for mid-market enterprises. Known in the industry as a natural storyteller, David has an unusual ability to make complex security concepts accessible to non-technical audiences—a skill that's served him well in fundraising and enterprise sales alike.\n\nBefore starting Beacon, David spent nearly a decade at Palo Alto Networks, where he rose from solutions engineer to director of product strategy. His time there gave him deep insight into the gaps in the market: enterprise tools were too expensive and complex for smaller companies, while SMB solutions lacked the sophistication needed to combat modern threats. This observation became the thesis behind [Beacon](companies/beacon-10).\n\nWang is often described as patient by colleagues and investors, a trait that manifests in his methodical approach to company building. He's not chasing hypergrowth at all costs. Instead, David prefers sustainable expansion, focusing on customer retention and product depth before scaling the sales team. This philosophy has occasionally put him at odds with more aggressive board members, but the results speak for themselves—Beacon boasts a net revenue retention rate above 130%.\n\nOn the personal side, David grew up in Vancouver before moving to the Bay Area for undergrad at Stanford. He studied computer science but minored in creative writing, which explains his storytelling instincts. He's married with two kids and is known for his weekend hiking trips around Marin County. Colleagues note he often returns from these hikes with new product ideas scribbled in a moleskin notebook.\n\nDavid speaks regularly at cybersecurity conferences, particularly RSA and Black Hat, where his talks tend to focus on the human element of security rather than purely technical deep-dives. He's built a modest but engaged following on LinkedIn through his weekly posts on security culture and leadership lessons. At 41, he's considered a rising voice in the cybersecurity founder community.",
-  "timeline": "- **2021-03-15** | David Wang officially incorporates Beacon and begins recruiting founding engineering team\n- **2021-09-22** | Closes $4.2M seed round led by Costanoa Ventures\n- **2022-04-10** | Beacon launches private beta with 12 design partners\n- **2022-11-08** | Delivers keynote at regional cybersecurity summit in San Jose on \"building security culture from day one\"\n- **2023-06-14** | [Beacon](companies/beacon-10) announces Series A of $18M, David featured in TechCrunch profile\n- **2023-12-01** | Hires first VP of Sales, begins expanding go-to-market motion\n- **2024-05-19** | Speaks at RSA Conference on incident response for resource-constrained teams\n- **2024-10-03** | Named to Business Insider's \"Founders to Watch in Cybersecurity\" list\n- **2025-02-27** | Beacon crosses 200 enterprise customers milestone\n- **2025-08-11** | David hosts private dinner for security founders in SF, begins informal founder network",
-  "_facts": {
-    "type": "person",
-    "slug": "people/david-wang-10",
-    "name": "David Wang",
-    "role": "founder",
-    "primary_affiliation": "companies/beacon-10",
-    "notable_traits": [
-      "storyteller",
-      "patient"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__david-zhang-83.json b/eval/data/world-v1/people__david-zhang-83.json
deleted file mode 100644
index 6a6531e37..000000000
--- a/eval/data/world-v1/people__david-zhang-83.json
+++ /dev/null
@@ -1,23 +0,0 @@
-{
-  "slug": "people/david-zhang-83",
-  "type": "person",
-  "title": "David Zhang",
-  "compiled_truth": "David Zhang is a partner at [Benchmark](companies/benchmark-3), one of Silicon Valley's most storied venture capital firms. Known for his demanding nature and an almost uncanny ability to weave narratives around the companies he backs, Zhang has carved out a reputation as both a kingmaker and a relentless operator.\n\nBefore joining Benchmark, David spent nearly a decade as an operator himself. He cut his teeth at a series of enterprise SaaS startups in the late 2010s, eventually rising to VP of Product at a mid-stage company that was acqui-hired by Salesforce. That experience—watching founders struggle to articulate their vision to investors—shaped his philosophy on storytelling. He's been quoted saying that \"the best founders don't just build products, they build worlds that others want to inhabit.\"\n\nAt Benchmark, Zhang led the firm's investment in [Gamma Labs](companies/gamma-labs-52), a developer tools company that's quietly becoming a darling among infrastructure engineers. He sits on the board there and is known for pushing the team hard on go-to-market execution. Some say too hard. But the results speak: Gamma Labs has tripled ARR in the past eighteen months.\n\nDavid also holds board observer seats at [Delta](companies/delta-3) and [Vector](companies/vector-6), both earlier-stage bets that reflect his interest in verticalized AI applications. He's less hands-on with these companies but remains a sounding board for their founders, particularly on fundraising strategy and narrative positioning.\n\nOutside of work, Zhang is a voracious reader of history and fiction alike. He credits his storytelling instincts to years of reading—everything from Thucydides to Ursula K. Le Guin. Colleagues describe him as intense, sometimes abrasive, but ultimately fair. He doesn't suffer fools, but he'll go to the mat for founders he beleives in.\n\nDavid lives in San Francisco with his wife and two kids. He ocasionally angel invests but keeps a low profile on those deals, preferring to let his Benchmark work speak for itself.",
-  "timeline": "- **2021-03-15** | David Zhang joins [Benchmark](companies/benchmark-3) as partner after leaving Salesforce\n- **2021-09-02** | Led Series A investment in [Gamma Labs](companies/gamma-labs-52), takes board seat\n- **2022-04-18** | Keynote at SaaStr Annual on \"Narrative as Competitive Advantage\"\n- **2022-11-30** | Participated in seed round for [Delta](companies/delta-3)\n- **2023-06-12** | Internal strategy session with Gamma Labs team on enterprise expansion\n- **2024-01-22** | Board observer role formalized at [Vector](companies/vector-6) following Series A\n- **2024-08-05** | Quoted in The Information piece about Benchmark's AI thesis\n- **2025-02-14** | Hosts founder dinner in SF, theme: \"building for the long arc\"\n- **2025-10-03** | Gamma Labs crosses $50M ARR milestone, Zhang celebrates publicly on X",
-  "_facts": {
-    "type": "person",
-    "slug": "people/david-zhang-83",
-    "name": "David Zhang",
-    "role": "partner",
-    "primary_affiliation": "companies/benchmark-3",
-    "secondary_affiliations": [
-      "companies/delta-3",
-      "companies/gamma-labs-52",
-      "companies/vector-6"
-    ],
-    "notable_traits": [
-      "storyteller",
-      "demanding"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__eric-lee-21.json b/eval/data/world-v1/people__eric-lee-21.json
deleted file mode 100644
index 756adf372..000000000
--- a/eval/data/world-v1/people__eric-lee-21.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/eric-lee-21",
-  "type": "person",
-  "title": "Eric Lee",
-  "compiled_truth": "Eric Lee is the founder of [Lucid](companies/lucid-21), a climate tech startup focused on industrial decarbonization. He's known in founder circles as a systems builder—someone who obsesses over architecting processes and infrastructure before scaling. This approach has made Lucid slower to market than some competitors, but the company's technical foundation is reportedly rock-solid.\n\nBefore starting [Lucid](companies/lucid-21), Eric spent six years at an enterprise software company where he led platform engineering. He's talked openly about how that experience shaped his views on technical debt and why he's so opinionated about building things right the first time. Some find his strong opinions abrasive; others see it as refreshing clarity in a space full of greenwashing and vaporware.\n\nEric grew up in Seattle and studied mechanical enginering at Stanford before pivoting to software. He's mentioned in interviews that the pivot came after realizing software could have more leverage on climate outcomes than hardware alone. That said, Lucid's approach blends both—they build software tools that optimize industrial processes, but they work closely with physical systems and manufacturing.\n\nHe's not a frequent conference speaker, preferring to stay heads-down on product. When he does appear publicly, he tends to be blunt about what's broken in climate tech funding and why most carbon accounting startups are \"solving the wrong problem.\" This has earned him a small but loyal following among technical founders who share his skepticism of hype cycles.\n\nEric is based in Oakland and runs a tight team of about 15 people. He's known for hiring slowly and valuing depth over breadth. Former colleagues describe him as intense but fair—someone who pushes hard but also takes feedback seriously. He's currently focused on expanding Lucid's pilot programs with several heavy industrial partners.",
-  "timeline": "- **2021-03-15** | Eric Lee incorporates [Lucid](companies/lucid-21) after leaving his platform engineering role\n- **2021-09-02** | Closes pre-seed round, mostly from climate-focused angels\n- **2022-04-18** | First pilot deployment with a cement manufacturer in Texas\n- **2022-11-30** | Publishes a widely-shared blog post criticizing carbon offset markets\n- **2023-06-12** | Seed round closes at $4.2M, led by a climate fund\n- **2023-10-05** | Hired third engineer after 8-month search process\n- **2024-02-22** | Speaks at Climate Tech Summit, calls out \"dashboard theater\" in the industry\n- **2024-08-14** | Signs LOI with major steel producer for expanded pilot\n- **2025-01-09** | Team hits 15 people after slow but steady hiring\n- **2025-05-20** | Begins Series A conversations with growth-stage investors",
-  "_facts": {
-    "type": "person",
-    "slug": "people/eric-lee-21",
-    "name": "Eric Lee",
-    "role": "founder",
-    "primary_affiliation": "companies/lucid-21",
-    "notable_traits": [
-      "systems builder",
-      "opinionated"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__eric-martinez-93.json b/eval/data/world-v1/people__eric-martinez-93.json
deleted file mode 100644
index 4f1770649..000000000
--- a/eval/data/world-v1/people__eric-martinez-93.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "people/eric-martinez-93",
-  "type": "person",
-  "title": "Eric Martinez",
-  "compiled_truth": "Eric Martinez is a partner at [NEA](companies/nea-13), one of the largest and most established venture capital firms in the world. He joined NEA in 2019 after a stint at Andreessen Horowitz where he focused primarily on enterprise software and developer tools. Eric has built a reputation as someone with strong opinions about markets—he's not afraid to pass on deals that others are chasing, and he's equally comfortable leading rounds in companies that the rest of Sand Hill Road hasn't discovered yet.\n\nMartinez is perhaps best known for his recruiting strength. Founders he's backed consistently cite his ability to help them land key executive hires, particularly in engineering and product roles. He maintains an unusually deep network of operators, many of whom he's stayed in touch with since his days at Stanford GSB. This network has proven especially valuable for portfolio companies scaling from Series A to B, a transition where many startups stumble on talent.\n\nHis current board seats include [Compass Labs](companies/compass-labs-61), a geospatial analytics company, and [Spire](companies/spire-46), which builds satellite-based data infrastructure. He was also an early investor in [Drift Labs](companies/drift-labs-81), though he's since rotated off that board. More recently, Eric led NEA's investment in [Pulse](companies/pulse-8), a healthcare AI startup focused on remote patient monitoring—a deal he sourced through a former colleague at a]h16z.\n\nEric tends to write long memos internally before committing to an investment, a habit some partners find tedious but which has helped NEA avoid several high-profile blowups. He's opinionated about market timing, often arguing that being two years early is functionally the same as being wrong. This perspective sometimes puts him at odds with more thesis-driven investors, but his track record speaks for itself. Martinez splits his time between San Francisco and NEA's Menlo Park office, though he's been spending more time in New York lately as the firm expands its east coast presence.",
-  "timeline": "- **2021-03-15** | Eric led NEA's Series A investment in [Compass Labs](companies/compass-labs-61), taking a board seat\n- **2021-09-02** | Spoke at TechCrunch Disrupt on the panel \"What VCs Actually Look For\"\n- **2022-04-18** | Helped [Spire](companies/spire-46) recruit their new VP of Engineering from Palantir\n- **2022-11-30** | Published essay on NEA blog arguing against \"spray and pray\" seed investing\n- **2023-06-12** | Led Series B for [Pulse](companies/pulse-8), $42M round at $180M valuation\n- **2023-10-05** | Rotated off [Drift Labs](companies/drift-labs-81) board after Series C lead took seat\n- **2024-02-20** | Internal memo on AI infrastructure opportunities circulated widely at NEA\n- **2024-08-14** | Joined Stanford GSB as a visiting lecturer for fall quarter\n- **2025-01-09** | Announced as keynote speaker for SaaStr Annual 2025\n- **2025-05-22** | Closed undisclosed investment in stealth robotics company",
-  "_facts": {
-    "type": "person",
-    "slug": "people/eric-martinez-93",
-    "name": "Eric Martinez",
-    "role": "partner",
-    "primary_affiliation": "companies/nea-13",
-    "secondary_affiliations": [
-      "companies/compass-labs-61",
-      "companies/spire-46",
-      "companies/drift-labs-81",
-      "companies/pulse-8"
-    ],
-    "notable_traits": [
-      "recruiting strength",
-      "opinionated"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__eric-miller-35.json b/eval/data/world-v1/people__eric-miller-35.json
deleted file mode 100644
index 73bb3ff3c..000000000
--- a/eval/data/world-v1/people__eric-miller-35.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/eric-miller-35",
-  "type": "person",
-  "title": "Eric Miller",
-  "compiled_truth": "Eric Miller is the founder of [Hatch](companies/hatch-35), an edtech startup focused on making early childhood learning more accessible through thoughtfully designed digital experiences. He's become known in founder circles for two things: shipping fast and having unusually good design taste for a technical founder.\n\nBefore starting Hatch, Eric spent three years at a larger edtech company where he grew frustrated with how slowly products moved and how little attention was paid to the actual user experience for kids. He left in early 2022 to build something different. The early versions of Hatch were scrappy—Eric essentially lived in Figma and code for six months straight, pushing updates sometimes multiple times per day.\n\nWhat sets Eric apart from other founders is his refusal to ship anything that feels mediocre. He's been known to delay features by weeks because the animations didn't feel right, or because the onboarding flow was confusing for a five-year-old. This attention to detail has earned [Hatch](companies/hatch-35) a loyal following among parents who are tired of apps that feel like afterthoughts.\n\nEric Miller is quiet in most settings but becomes animated when discussing interaction design or learning theory. He reads obsessively—developmental psychology papers, old Nintendo design documents, Montessori philosophy. His Twitter presence is minimal, mostly just occasional screenshots of works-in-progress that get shared widely in design communites.\n\nHe raised a seed round in late 2022, though he's been cagey about the exact amount. Investors describe him as stubborn in the best way—he knows what he wants to build and won't be pushed toward generic growth tactics. Miller has said publicly that he'd rather Hatch be used by 100,000 families who love it than a million who forget about it.\n\nAt 31, Eric lives in Portland and works mostly from home. He's mentioned in interviews that becoming a parent himself in 2023 changed how he thinks about the product. The stakes feel more real now, he says.",
-  "timeline": "- **2022-02-14** | Eric Miller officially incorporates Hatch after leaving his previous role\n- **2022-08-03** | First public beta of [Hatch](companies/hatch-35) launches with 500 families\n- **2022-11-19** | Closes seed round, terms undisclosed but rumored around $2.1M\n- **2023-03-22** | Eric speaks at EdTech Builders Summit on \"Designing for Pre-Readers\"\n- **2023-07-08** | Hatch crosses 25,000 active users, Eric posts rare celebratory tweet\n- **2023-09-15** | First child born, takes two weeks off—longest break since founding\n- **2024-01-29** | Major app redesign ships after four months of iteration\n- **2024-06-11** | Featured in Fast Company's \"Most Creative People in Edtech\" list\n- **2024-11-03** | Begins advising two early-stage founders on product design\n- **2025-02-18** | Hints at upcoming expansion into early literacy tools for [Hatch](companies/hatch-35)",
-  "_facts": {
-    "type": "person",
-    "slug": "people/eric-miller-35",
-    "name": "Eric Miller",
-    "role": "founder",
-    "primary_affiliation": "companies/hatch-35",
-    "notable_traits": [
-      "fast-shipping",
-      "design taste"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__fiona-moore-88.json b/eval/data/world-v1/people__fiona-moore-88.json
deleted file mode 100644
index 4b55b3c9b..000000000
--- a/eval/data/world-v1/people__fiona-moore-88.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "people/fiona-moore-88",
-  "type": "person",
-  "title": "Fiona Moore",
-  "compiled_truth": "Fiona Moore is a partner at [Khosla Ventures](companies/khosla-ventures-8), where she's built a reputation as one of the more opinionated voices in climate and deep tech investing. She joined Khosla in 2019 after a stint in corporate strategy at a major energy company, and has since led or co-led investments across energy storage, synthetic biology, and developer tools.\n\nMoore is known for her long-term thinking—she's publicly stated that she evaluates founders on whether they can articulate a 15-year vision, not just a Series B strategy. This has made her a polarizing figure in some circles; founders either love her directness or find her feedback too blunt. But her portfolio speaks for itself. She led Khosla's investment in [Prism Labs](companies/prism-labs-93), an early-stage company working on next-gen photovoltaic materials, and sits on the board of [Spire Labs](companies/spire-labs-96), which is building modular carbon capture units for industrial applications.\n\nBeyond her primary work, Fiona maintains board observer seats at [Lucid](companies/lucid-21) and [Drift](companies/drift-31), both of which she championed internally before they became consensus picks. She's also an advisor to [Vox](companies/vox-25), though that relationship is more informal—mostly product feedback and occasional intros.\n\nBefore venture, Moore studied chemical enginering at MIT and spent two years at McKinsey, which she rarely mentions. She's more likely to talk about the summer she spent working on a geothermal project in Iceland, which she credits with shaping her investment thesis around energy infrastructure.\n\nFiona speaks frequently at climate-focused conferences and has a somewhat cultish following on Twitter, where she posts threads on energy policy and occasionaly dunks on \"tourist VCs\" who she feels don't understand hardware timelines. She lives in San Francisco with her partner and two rescue dogs. Close colleagues describe her as intense but deeply loyal—the kind of investor who will take a 9pm call from a struggling founder without hesitation.",
-  "timeline": "- **2021-03-12** | Fiona Moore joins the board of [Spire Labs](companies/spire-labs-96) following their Series A\n- **2021-09-08** | Keynote at Climate Tech Summit in Denver on \"Patient Capital for Hard Problems\"\n- **2022-02-14** | Leads [Khosla Ventures](companies/khosla-ventures-8) seed investment in [Prism Labs](companies/prism-labs-93)\n- **2022-11-03** | Named to Forbes Midas List honorable mentions for climate investing\n- **2023-04-21** | Hosts private dinner with [Lucid](companies/lucid-21) and [Drift](companies/drift-31) founders in SF\n- **2023-08-30** | Publishes widely-shared memo on hardware startup timelines and investor patience\n- **2024-01-17** | Joins [Vox](companies/vox-25) as informal product advisor\n- **2024-06-09** | Speaks at Khosla annual LP meeting on deep tech portfolio performance\n- **2025-02-22** | [Prism Labs](companies/prism-labs-93) announces Series B; Moore leads follow-on\n- **2025-11-14** | Internal promotion to senior partner at Khosla Ventures",
-  "_facts": {
-    "type": "person",
-    "slug": "people/fiona-moore-88",
-    "name": "Fiona Moore",
-    "role": "partner",
-    "primary_affiliation": "companies/khosla-ventures-8",
-    "secondary_affiliations": [
-      "companies/vox-25",
-      "companies/spire-labs-96",
-      "companies/lucid-21",
-      "companies/prism-labs-93",
-      "companies/drift-31"
-    ],
-    "notable_traits": [
-      "long-term thinker",
-      "opinionated"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__frank-hernandez-31.json b/eval/data/world-v1/people__frank-hernandez-31.json
deleted file mode 100644
index 52ed4ffa4..000000000
--- a/eval/data/world-v1/people__frank-hernandez-31.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/frank-hernandez-31",
-  "type": "person",
-  "title": "Frank Hernandez",
-  "compiled_truth": "Frank Hernandez is the founder of [Drift](companies/drift-31), a developer tools company focused on building infrastructure for real-time collaboration in code editors. Before starting Drift, Frank spent six years at Stripe where he led the team responsible for their internal developer productivity tools. His work there became somewhat legendary among infra nerds—he was the architect behind their hot-reload system that cut local dev cycle times by 80%.\n\nHernandez is known as a first-principles thinker who refuses to accept conventional wisdom about how things should work. When he started [Drift](companies/drift-31), he threw out the entire premise of how multiplayer coding should function. Instead of bolting collaboration onto existing editors, he rebuilt the editing model from scratch around the assumption that code is inherently social. This systems builder mentality pervades everything about the company.\n\nFrank grew up in Austin, Texas and studied computer science at UT Austin before dropping out his junior year to join a YC startup that eventually failed. He doesn't talk about that experience much, but colleagues say it taught him the importance of distribution over pure product quality. \"You can build the most elegant system in the world,\" he's said in interviews, \"and it means nothing if developers dont actually adopt it.\"\n\nHis management style is hands-off but demanding. He hires senior engineers almost exclusively and expects them to own entire problem spaces autonomously. Weekly syncs at Drift are famously short—fifteen minutes max. Frank belives that if you need more time than that, you haven't thought clearly enough about what you're working on.\n\nOutside of work, Hernandez is an avid rock climber and often draws parallels between climbing routes and system design. He's also known to be an early riser, often sending Slack messages at 5am that he insists nobody needs to respond to until normal hours. Frank remains deeply technical despite his CEO role, still committing code to Drift's core sync engine on a weekly basis.",
-  "timeline": "- **2021-03-15** | Frank leaves Stripe after six years to start working on what would become [Drift](companies/drift-31)\n- **2021-08-22** | Incorporates Drift and closes a $2.1M pre-seed round from angels including former Stripe colleagues\n- **2022-01-10** | Ships first private alpha of Drift to 50 hand-picked developer teams\n- **2022-09-08** | Speaks at Strange Loop about \"Why CRDT's aren't enough for real-time code collaboration\"\n- **2023-04-17** | [Drift](companies/drift-31) announces $18M Series A, Frank quoted extensively in TechCrunch coverage\n- **2023-11-02** | Publishes influential blog post \"The Latency Tax\" about developer tool performance\n- **2024-06-20** | Keynote at GitHub Universe, demos Drift's VS Code integration to mass audience\n- **2024-12-03** | Named to Forbes 30 Under 30 (enterprise technology category)\n- **2025-05-14** | Internal all-hands where Frank announces pivot toward AI-assisted pair programming features\n- **2025-09-28** | Drift hits 100k monthly active developers, Frank celebrates with team offsite in Colorado",
-  "_facts": {
-    "type": "person",
-    "slug": "people/frank-hernandez-31",
-    "name": "Frank Hernandez",
-    "role": "founder",
-    "primary_affiliation": "companies/drift-31",
-    "notable_traits": [
-      "first-principles thinker",
-      "systems builder"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__helen-johnson-32.json b/eval/data/world-v1/people__helen-johnson-32.json
deleted file mode 100644
index b1df8a2f2..000000000
--- a/eval/data/world-v1/people__helen-johnson-32.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/helen-johnson-32",
-  "type": "person",
-  "title": "Helen Johnson",
-  "compiled_truth": "Helen Johnson is the founder of [Echo](companies/echo-32), a robotics company thats been making waves in the industrial automation space. She's known for being intensely product-obsessed, often spending late nights in the lab tweaking hardware prototypes herself rather than delegating to her engineering team. Colleagues describe her as analytical to a fault—she won't greenlight a feature unless the data backs it up completely.\n\nBefore starting Echo, Helen spent six years at Boston Dynamics where she led a small skunkworks team focused on warehouse logistics robots. That experience shaped her philosophy around building machines that actually work in messy, real-world environments rather than just performing well in controlled demos. She left in 2021 after growing frustrated with the pace of commercialization.\n\nJohnson raised a $12M seed round for [Echo](companies/echo-32) in early 2022, primarily from deep-tech focused investors who bought into her vision of modular robotic systems for mid-sized manufacturers. The thesis was simple: most robotics companies target either massive enterprises or consumers, leaving a gap in the market for companies doing $50-200M in revenue who need automation but cant afford custom solutions.\n\nHer management style is polarizing. Some engineers thrive under her direct, no-nonsense feedback. Others have found it exhausting. Turnover in the first two years was higher than ideal, though the team has stabilized since bringing on a strong VP of Engineering in 2023. Helen has been open about learning to delegate more effectivley as the company scales.\n\nShe holds a PhD in mechanical engineering from MIT and a BS from Georgia Tech. Outside of work, she's an avid rock climber and has mentioned in interviews that climbing taught her to think through problems methodically while staying calm under pressure. She's based in Boston but travels frequently to Echo's manufacturing partner in Ohio.\n\nHelen doesn't do much public speaking, preferring to let the product speak for itself. When she does appear at conferences, her presentations are notably data-dense and light on hype—a refreshing contrast in the robotics world.",
-  "timeline": "- **2021-03-15** | Helen leaves Boston Dynamics to start working on Echo concept full-time\n- **2022-01-20** | Closes $12M seed round for [Echo](companies/echo-32) led by Construct Capital\n- **2022-09-08** | First prototype unit deployed at pilot customer facility in Toledo\n- **2023-04-12** | Hires VP of Engineering Marcus Webb from Fetch Robotics\n- **2023-11-30** | Echo ships its 50th unit, hits $4M ARR milestone\n- **2024-03-22** | Keynote presentation at Automate 2024 in Chicago, demos new gripper system\n- **2024-08-15** | Announces Series A raise, terms undisclosed but rumored around $30M\n- **2025-02-10** | Featured in IEEE Spectrum profile on next-gen manufacturing robotics\n- **2025-09-18** | [Echo](companies/echo-32) expands to second manufacturing facility",
-  "_facts": {
-    "type": "person",
-    "slug": "people/helen-johnson-32",
-    "name": "Helen Johnson",
-    "role": "founder",
-    "primary_affiliation": "companies/echo-32",
-    "notable_traits": [
-      "product-obsessed",
-      "analytical"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__helen-martinez-87.json b/eval/data/world-v1/people__helen-martinez-87.json
deleted file mode 100644
index a16eac4df..000000000
--- a/eval/data/world-v1/people__helen-martinez-87.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "people/helen-martinez-87",
-  "type": "person",
-  "title": "Helen Martinez",
-  "compiled_truth": "Helen Martinez is a partner at [Index Ventures](companies/index-ventures-7), where she's built a reputation for backing founders who ship fast and figure out distribution early. Her portfolio skews toward developer tools and infrastructure plays, but she's increasingly spending time in vertical SaaS and AI-native applications.\n\nBefore joining Index, Helen spent six years on the operating side—first at a Series B fintech that got aquired by Stripe, then leading growth at a productivity startup that flamed out spectacularly in 2019. She doesn't talk about the failure much, but it clearly shaped her investment thesis: she's allergic to companies that can't articulate their go-to-market motion by the seed stage.\n\nHelen sits on the boards of [Acme Labs](companies/acme-labs-50) and [Nexus Labs](companies/nexus-labs-91), both of which she led from seed. She's also an observer or advisor at [Ranger](companies/ranger-22), [Mosaic](companies/mosaic-14), and [Echo](companies/echo-32). Her board style is hands-on without being overbearing—founders describe her as \"the person who actually reads the weekly update and responds with something useful.\"\n\nMartinez is known for moving quickly. She's reportedly made investment decisions in under 48 hours when she sees a team that matches her pattern: technical founders with unusual distribution insight. She's less interested in pure research plays or deep tech that requires years of R&D before hitting market.\n\nOn the conference circuit, Helen tends to skip the main stage and instead hosts smaller dinners for portfolio founders. She's been vocal about the importance of founder mental health and has pushed Index to expand its support resources in that area. Some partners find her intensity exhausting; founders mostly find it reassuring.\n\nShe splits time between San Francisco and London, though lately she's been spending more weeks in Europe as Index doubles down on its EU presence. Helen is active on Twitter but keeps her LinkedIn sparse—she says she evaluates founders partly on whether they cold-email her directly instead of trying to get warm intros.",
-  "timeline": "- **2021-03-15** | Led seed round for [Acme Labs](companies/acme-labs-50), her first board seat at Index\n- **2021-09-02** | Spoke at SaaStr Annual on \"Why GTM Belongs in Your Seed Deck\"\n- **2022-01-20** | Joined board of [Nexus Labs](companies/nexus-labs-91) after preempting their Series A\n- **2022-08-11** | Published internal memo on AI investment thesis that leaked to Twitter\n- **2023-04-07** | Became advisor to [Ranger](companies/ranger-22) through Index's scout program\n- **2023-11-30** | Hosted first annual portfolio founder retreat in Lisbon\n- **2024-02-14** | Led [Mosaic](companies/mosaic-14) Series A alongside Sequoia\n- **2024-09-19** | Promoted to senior partner at [Index Ventures](companies/index-ventures-7)\n- **2025-03-22** | Added [Echo](companies/echo-32) to portfolio via seed extension\n- **2025-12-01** | Named to Forbes Midas List for first time",
-  "_facts": {
-    "type": "person",
-    "slug": "people/helen-martinez-87",
-    "name": "Helen Martinez",
-    "role": "partner",
-    "primary_affiliation": "companies/index-ventures-7",
-    "secondary_affiliations": [
-      "companies/acme-labs-50",
-      "companies/nexus-labs-91",
-      "companies/ranger-22",
-      "companies/mosaic-14",
-      "companies/echo-32"
-    ],
-    "notable_traits": [
-      "GTM-heavy",
-      "fast-shipping"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__henry-johnson-12.json b/eval/data/world-v1/people__henry-johnson-12.json
deleted file mode 100644
index f0d9656af..000000000
--- a/eval/data/world-v1/people__henry-johnson-12.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/henry-johnson-12",
-  "type": "person",
-  "title": "Henry Johnson",
-  "compiled_truth": "Henry Johnson is the founder and CEO of [Lumen](companies/lumen-12), a biotech company focused on next-generation metabolic sensing platforms. Before starting Lumen, Henry spent nearly a decade in academic research at MIT, where he completed his PhD in biomedical engineering and led a lab focused on non-invasive biosensor development. His technical background runs deep—colleagues describe him as someone who can hold his own in conversations ranging from electrode chemistry to firmware optimization.\n\nWhat sets Henry apart from many first-time founders is his surprising fluency in fundraising. He closed Lumen's seed round in under six weeks, a feat that caught the attention of several Boston-area VCs who'd initially passed. His pitch decks are unusually technical, packed with data that most investors wouldn't fully parse, but he delivers them with a clarity that builds confidence. One investor noted that Johnson \"makes you feel smarter for listening to him,\" which is rare in biotech pitches that often devolve into jargon.\n\nHenry is known for being direct, sometimes to a fault. He doesn't sugarcoat timelines or overstate clinical readiness, which has occasionally frustrated board members expecting more optimistic projections. But this honesty has also earned him a reputation for reliability—when he says something will ship, it usually does. His management style leans technical; he still reviews code commits and prototype specs weekly, even as [Lumen](companies/lumen-12) has grown past 30 employees.\n\nOutside of work, Johnson keeps a low profile. He's not on the conference circuit and rarely does press interviews. When he does speak publicly, it's typically at niche biotech meetups or university seminars. He's mentioned in passing that he prefers \"building to talking about building,\" a sentiment that tracks with his general demeanor. Freinds from grad school say he was always this way—obsessive about the work, skeptical of hype. Now in his late thirties, Henry seems focused on proving that Lumen can become a category-defining company in metabolic health, one careful experiment at a time.",
-  "timeline": "- **2021-03-15** | Henry Johnson incorporates [Lumen](companies/lumen-12) as a Delaware C-corp, initially self-funded\n- **2021-09-02** | Closes $2.4M seed round led by Arch Ventures in just six weeks\n- **2022-01-20** | Hires first full-time engineer, begins prototype development in Cambridge lab\n- **2022-08-11** | Presents early sensor data at MIT biosensing symposium, generates academic buzz\n- **2023-02-28** | Lumen completes Series A, raising $18M; Henry leads investor negotiations directly\n- **2023-07-14** | First pilot units shipped to three clinical research partners\n- **2024-01-09** | Keynote at Biotech Founders Summit in San Francisco, rare public appearance\n- **2024-06-22** | Team grows to 30+ employees; Henry still reviews weekly prototype specs\n- **2025-03-30** | Announces partnership with major diabetes research consortium\n- **2025-11-18** | Featured in Nature Biotechnology profile on non-invasive sensing pioneers",
-  "_facts": {
-    "type": "person",
-    "slug": "people/henry-johnson-12",
-    "name": "Henry Johnson",
-    "role": "founder",
-    "primary_affiliation": "companies/lumen-12",
-    "notable_traits": [
-      "technical depth",
-      "fundraising-savvy"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__ian-davis-33.json b/eval/data/world-v1/people__ian-davis-33.json
deleted file mode 100644
index d61b2d9e8..000000000
--- a/eval/data/world-v1/people__ian-davis-33.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/ian-davis-33",
-  "type": "person",
-  "title": "Ian Davis",
-  "compiled_truth": "Ian Davis is the founder of [Foundry](companies/foundry-33), an AI applications company he started after spending nearly a decade building infrastructure at scale. Before Foundry, Davis worked at several enterprise software companies where he developed a reputation as someone who could take messy, undefined problems and turn them into working systems. Colleagues from those early days describe him as relentless—the kind of person who would rewrite an entire codebase over a weekend if he thought the architecture was wrong.\n\nDavis launched [Foundry](companies/foundry-33) in late 2022, initially focused on building internal AI tools for mid-market companies that couldn't afford custom solutions but had outgrown basic SaaS offerings. The company has since expanded into vertical-specific applications, particularly in logistics and healthcare operations. Ian tends to avoid the spotlight, rarely speaking at conferences or doing press, though he's known in founder circles for being exceptionally demanding of his team. One former engineer described working with him as \"exhausting but educational\"—he expects first-principles thinking and has little patience for surface-level answers.\n\nHis approach to building Foundry reflects this intensity. The company ships quickly but maintains unusually high standards for code quality and system reliability. Davis himself still reviews critical pull requests and has been known to block releases over what others might consider minor issues. This has created some tension as the team has grown, but it's also contributed to Foundry's reputation for building products that actually work in production enviroments.\n\nOutside of Foundry, Davis is a private person. He doesn't maintain much of a social media presence and keeps his personal life seperate from his professional one. Those who know him say he's deeply curious about complex systems of all kinds—not just software, but economics, biology, and urban planning. He reads widely and often connects ideas across domains in ways that inform his product thinking. At 38, he's still very much in building mode, with no apparent interest in transitioning to a more advisory role anytime soon.",
-  "timeline": "- **2022-11-14** | Ian Davis incorporates Foundry, begins working on initial prototypes from his apartment in Austin\n- **2023-02-08** | Closes a $2.4M seed round led by a small syndicate of enterprise software operators\n- **2023-06-22** | Hires first three engineers, all former colleagues from his infrastructure days\n- **2023-09-15** | Ships v1 of Foundry's logistics optimization tool to two pilot customers\n- **2024-01-29** | Gives rare talk at a private founder dinner about building AI apps that survive contact with real data\n- **2024-05-03** | [Foundry](companies/foundry-33) crosses $1M ARR, team grows to eleven people\n- **2024-08-17** | Publicly criticizes a competitor's approach to AI safety on a podcast appearance, generates minor controversy\n- **2025-01-11** | Begins exploring healthcare operations vertical after conversations with hospital system CTO\n- **2025-04-28** | Raises Series A, terms undisclosed but rumored around $12M\n- **2025-09-06** | Featured in a profile about demanding founder archetypes in a niche tech newsletter",
-  "_facts": {
-    "type": "person",
-    "slug": "people/ian-davis-33",
-    "name": "Ian Davis",
-    "role": "founder",
-    "primary_affiliation": "companies/foundry-33",
-    "notable_traits": [
-      "systems builder",
-      "demanding"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__iris-lee-82.json b/eval/data/world-v1/people__iris-lee-82.json
deleted file mode 100644
index 07bb89d21..000000000
--- a/eval/data/world-v1/people__iris-lee-82.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "people/iris-lee-82",
-  "type": "person",
-  "title": "Iris Lee",
-  "compiled_truth": "Iris Lee is a partner at [Andreessen Horowitz](companies/andreessen-horowitz-2), where she focuses primarily on enterprise infrastructure and developer tools. Known for being deeply product-obsessed, she spends an unusual amount of time actually using the products her portfolio companies build—often filing bug reports and feature requests alongside regular users.\n\nBefore joining a16z, Iris cut her teeth in product management at several early-stage startups, which gives her a practitioners lens that founders seem to appreciate. She's particularly fundraising-savvy, having helped numerous companies navigate the complexities of Series A through growth rounds. Her network runs deep across both coasts, and she's developed a reputation for being able to close competitive deals by moving fast and offering genuine strategic value beyond just capital.\n\nIris currently serves on the boards of several promising companies including [Vector Labs](companies/vector-labs-56), [Epsilon Labs](companies/epsilon-labs-54), and [Cipher Labs](companies/cipher-labs-63). She also maintains board observer seats at [Lattice Labs](companies/lattice-labs-89) and [Zenith Labs](companies/zenith-labs-77), both of which she led seed investments in before her promotion to partner. Her portfolio has a clear throughline: technical teams building picks-and-shovels infrastructure that other developers rely on.\n\nLee is originally from Seattle and studied computer science at Stanford before dropping out of a PhD program at MIT to join a YC company that eventually got acquihired by Dropbox. She spent three years there working on core sync infrastructure before making the jump to venture. Colleagues describe her as intensely analytical but surprisingly warm in person—the type who remembers your kids' names and follows up on things you mentioned in passing months ago.\n\nShe's become a regular presence on the conference circuit, frequently speaking about the evolution of developer experience and why she belives the next wave of infrastructure companies will be built around AI-native workflows. Her newsletter, \"The Build,\" has quietly amassed about 40k subscribers, mostly engineering leaders and founders. Iris remains one of the more hands-on partners at the firm, often pulling all-nighters alongside portfolio company teams during critical launches.",
-  "timeline": "- **2021-03-15** | Joined [Andreessen Horowitz](companies/andreessen-horowitz-2) as a deal partner, initially covering infrastructure deals\n- **2021-09-22** | Led seed round for [Lattice Labs](companies/lattice-labs-89), her first solo investment at the firm\n- **2022-02-10** | Sourced and closed Series A for [Epsilon Labs](companies/epsilon-labs-54) in a competitive process\n- **2022-08-04** | Promoted to full partner at a16z after strong portfolio performance\n- **2023-01-18** | Joined board of [Vector Labs](companies/vector-labs-56) following their Series B\n- **2023-06-30** | Gave keynote at DevCon SF on \"Infrastructure for the AI Era\"\n- **2023-11-12** | Led Series A for [Cipher Labs](companies/cipher-labs-63), takes board seat\n- **2024-04-25** | [Zenith Labs](companies/zenith-labs-77) announces $28M Series A, Iris maintains observer seat from seed\n- **2025-02-08** | Published widely-shared piece on enterprise AI adoption patterns\n- **2025-09-14** | Named to Forbes Midas List for first time, ranked #47",
-  "_facts": {
-    "type": "person",
-    "slug": "people/iris-lee-82",
-    "name": "Iris Lee",
-    "role": "partner",
-    "primary_affiliation": "companies/andreessen-horowitz-2",
-    "secondary_affiliations": [
-      "companies/vector-labs-56",
-      "companies/epsilon-labs-54",
-      "companies/cipher-labs-63",
-      "companies/lattice-labs-89",
-      "companies/zenith-labs-77"
-    ],
-    "notable_traits": [
-      "product-obsessed",
-      "fundraising-savvy"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__jack-davis-89.json b/eval/data/world-v1/people__jack-davis-89.json
deleted file mode 100644
index c1f01abe7..000000000
--- a/eval/data/world-v1/people__jack-davis-89.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "people/jack-davis-89",
-  "type": "person",
-  "title": "Jack Davis",
-  "compiled_truth": "Jack Davis is a partner at [Floodgate](companies/floodgate-9), one of the more established seed-stage venture capital firms in the Valley. He joined Floodgate in 2019 after a brief stint operating at a Series B startup that ultimately got acquried by Salesforce. Before that, Davis cut his teeth at a boutique advisory firm helping founders navigate early fundraising — an experience that shapes how he approaches deal sourcing today.\n\nDavis is known primarily for two things: his recruiting strength and his demanding nature. On the recruiting front, he's built a reputation for helping portfolio companies land critical early hires, particularly in engineering and product. Founders often cite his network as a key reason they chose Floodgate over competing term sheets. He maintains close relationships with executive recruiters and has a habit of personally vetting candidates for his portfolio companies' first ten hires. This hands-on approach has made him invaluable to companies like [Lumen Labs](companies/lumen-labs-62) and [Drift](companies/drift-31), where early team composition was make-or-break.\n\nThe demanding part is... well, it's a double-edged sword. Jack expects weekly updates, detailed metrics, and honest assessments of what's not working. Some founders thrive under this scrutiny; others find it exhausting. He's been known to push back hard in board meetings, questioning assumptions and forcing teams to defend their roadmaps. One founder described him as \"the coach you hate during practice but love when you win.\"\n\nHis current board seats include [Beta Labs](companies/beta-labs-51), [Iris](companies/iris-36), and [Quasar](companies/quasar-44). Across these companies, you see a clear pattern: developer tools, infrastructure, and B2B SaaS. Davis avoids consumer plays almost entirely — he's said publicly that he doesn't have the intuition for viral growth loops and prefers businesses where he can model the sales motion.\n\nOutside of Floodgate, Jack keeps a relatively low profile. He doesn't tweet much, rarely speaks at conferences, and seems to prefer the work of company-building over personal brand building. Married with two kids, lives somewhere in Menlo Park. Runs marathons ocasionally.",
-  "timeline": "- **2021-03-15** | Led seed round for [Lumen Labs](companies/lumen-labs-62), first check into the company\n- **2021-09-22** | Introduced [Drift](companies/drift-31) founders to their first VP of Engineering hire\n- **2022-01-10** | Joined board of [Beta Labs](companies/beta-labs-51) following Series A\n- **2022-07-08** | Spoke at internal Floodgate offsite on \"post-term sheet value-add\"\n- **2023-02-14** | Led pre-seed investment into [Iris](companies/iris-36)\n- **2023-11-30** | Participated in [Quasar](companies/quasar-44) seed extension round\n- **2024-04-19** | Hosted private dinner for infrastructure founders in SF\n- **2024-09-05** | Helped [Lumen Labs](companies/lumen-labs-62) close Series A, brought in co-lead from Sequoia\n- **2025-01-22** | Promoted to senior partner at [Floodgate](companies/floodgate-9)",
-  "_facts": {
-    "type": "person",
-    "slug": "people/jack-davis-89",
-    "name": "Jack Davis",
-    "role": "partner",
-    "primary_affiliation": "companies/floodgate-9",
-    "secondary_affiliations": [
-      "companies/lumen-labs-62",
-      "companies/drift-31",
-      "companies/beta-labs-51",
-      "companies/iris-36",
-      "companies/quasar-44"
-    ],
-    "notable_traits": [
-      "recruiting strength",
-      "demanding"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__julia-davis-86.json b/eval/data/world-v1/people__julia-davis-86.json
deleted file mode 100644
index 6d57680c1..000000000
--- a/eval/data/world-v1/people__julia-davis-86.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "people/julia-davis-86",
-  "type": "person",
-  "title": "Julia Davis",
-  "compiled_truth": "Julia Davis is a partner at [Lightspeed](companies/lightspeed-6), where she's built a reputation as one of the more operationally-minded investors in the firm's growth practice. Before joining Lightspeed in 2021, she spent nearly a decade in go-to-market roles at various startups, most recently as VP of Revenue at a Series C infrastructure company that was acquried by Datadog.\n\nHer investing thesis centers on what she calls \"systems-first GTM\" — the idea that the best companies treat their sales and marketing motions as engineering problems rather than art. This perspective has shaped her portfolio, which skews heavily toward developer tools and B2B infrastructure plays. She led Lightspeed's investment in [Quasar Labs](companies/quasar-labs-94) and sits on the board there, helping them build out their enterprise sales function from scratch.\n\nJulia also serves as a board observer at [Orbit](companies/orbit-42) and [Helix Labs](companies/helix-labs-59), both of which came through her network of technical founders. She's known for being deeply hands-on in the first 18 months post-investment, often spending two days a week with portfolio companies during critical scaling phases. Some founders find this intensity overwhelming; others credit her with saving them from costly GTM mistakes.\n\nBeyond her primary board seats, Davis maintains advisory relationships with [Echo](companies/echo-32) and [Mantle Labs](companies/mantle-labs-66). The Echo relationship is particularly notable — she passed on the deal initially but stayed close to the founders and eventually helped them restructure their pricing model, which she now points to as a case study in how VCs can add value without writing checks.\n\nShe's a systems builder at heart. Her internal playbooks for sales hiring and comp structures get passed around founder Slack groups regularly. Speaks frequently at SaaStr and other GTM-focused conferences. Lives in San Francisco with her partner and two kids, though she's on the road probably 40% of the time visiting portfolio companies.",
-  "timeline": "- **2021-03-15** | Joined [Lightspeed](companies/lightspeed-6) as a partner, transitioning from operating roles\n- **2021-09-22** | Led Series A investment in [Quasar Labs](companies/quasar-labs-94), her first deal at the firm\n- **2022-04-08** | Published influential blog post on \"systems-first GTM\" that got widely circulated in founder communities\n- **2022-11-30** | Joined board of [Helix Labs](companies/helix-labs-59) following their Series B\n- **2023-02-14** | Keynote at SaaStr Annual on building repeatable sales processes for technical products\n- **2023-08-19** | Began advisory relationship with [Echo](companies/echo-32) after passing on their round\n- **2024-01-11** | Helped [Orbit](companies/orbit-42) hire their first VP of Sales, took board observer seat\n- **2024-06-03** | Started working with [Mantle Labs](companies/mantle-labs-66) on pricing strategy\n- **2025-02-27** | Named to Forbes Midas List for first time\n- **2025-09-15** | Led [Quasar Labs](companies/quasar-labs-94) Series C at $180M valuation",
-  "_facts": {
-    "type": "person",
-    "slug": "people/julia-davis-86",
-    "name": "Julia Davis",
-    "role": "partner",
-    "primary_affiliation": "companies/lightspeed-6",
-    "secondary_affiliations": [
-      "companies/quasar-labs-94",
-      "companies/orbit-42",
-      "companies/helix-labs-59",
-      "companies/echo-32",
-      "companies/mantle-labs-66"
-    ],
-    "notable_traits": [
-      "GTM-heavy",
-      "systems builder"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__julia-johnson-114.json b/eval/data/world-v1/people__julia-johnson-114.json
deleted file mode 100644
index 11e0ca486..000000000
--- a/eval/data/world-v1/people__julia-johnson-114.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "people/julia-johnson-114",
-  "type": "person",
-  "title": "Julia Johnson",
-  "compiled_truth": "Julia Johnson is a cybersecurity engineer at [Epsilon](companies/epsilon-4), where she's built a reputation as one of the most patient and methodical threat analysts on the team. Colleagues describe her as someone who can stare at log files for hours without losing focus, picking out anomalies that others miss entirely. Her sharp pattern-matching abilities have made her invaluable during incident response situations.\n\nJulia joined [Epsilon](companies/epsilon-4) in early 2022 after spending three years at a smaller security consultancy that focused on penetration testing for fintech clients. The transition to Epsilon's enterprise-focused cybersecurity work was a natural fit. She'd grown tired of the constant client rotation and wanted to dig deeper into long-term defensive architecture.\n\nAt Epsilon, Johnson primarily works on threat detection systems and has contributed significantly to their behavioral analysis tooling. She's the kind of engineer who prefers to understand a problem completley before proposing solutions, which can sometimes frustrate faster-moving team members. But her patience pays off—her code tends to ship with fewer bugs and her threat assessments rarely miss the mark.\n\nShe's known internally for creating what the team calls the \"Julia Checklist,\" an exhaustive set of verification steps for validating potential security incidents. What started as her personal workflow document became official team protocol after it caught two false positives that would have triggered expensive incident responses.\n\nOutside of her core work, Julia has mentored several junior engineers and occassionally presents at internal knowledge-sharing sessions. She's not one for conference talks or public visibility, preferring to let her work speak for itself. Her manager has pushed her toward more leadership responsibilities, but she's been hesitant to move away from hands-on technical work.\n\nJulia holds a BS in Computer Science from Georgia Tech and picked up her security specialization through a combination of certifications and self-directed learning. She's active in a few private security research communities but keeps a low public profile.",
-  "timeline": "- **2021-03-15** | Completed OSCP certification while working at previous consultancy role\n- **2022-01-10** | Joined [Epsilon](companies/epsilon-4) as a security engineer on the threat detection team\n- **2022-09-22** | Led incident response for major client breach attempt, identified attack vector within 4 hours\n- **2023-02-14** | Published internal \"Julia Checklist\" documentation, later adopted as team standard\n- **2023-08-30** | Promoted to Senior Security Engineer at Epsilon\n- **2024-01-18** | Presented on behavioral anomaly detection at Epsilon's internal tech summit\n- **2024-06-05** | Began mentoring two junior engineers joining the cybersecurity team\n- **2024-11-12** | Contributed core detection logic to Epsilon's v3.0 threat analysis platform\n- **2025-04-08** | Recognized in company all-hands for catching sophisticated supply chain attack pattern",
-  "_facts": {
-    "type": "person",
-    "slug": "people/julia-johnson-114",
-    "name": "Julia Johnson",
-    "role": "engineer",
-    "primary_affiliation": "companies/epsilon-4",
-    "secondary_affiliations": [],
-    "notable_traits": [
-      "patient",
-      "sharp pattern matcher"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__kate-lopez-99.json b/eval/data/world-v1/people__kate-lopez-99.json
deleted file mode 100644
index 57d1ce26e..000000000
--- a/eval/data/world-v1/people__kate-lopez-99.json
+++ /dev/null
@@ -1,23 +0,0 @@
-{
-  "slug": "people/kate-lopez-99",
-  "type": "person",
-  "title": "Kate Lopez",
-  "compiled_truth": "Kate Lopez is a partner at [Greylock II](companies/greylock-ii-19), one of the more actively deploying venture firms in the enterprise infrastructure space. She joined Greylock in 2022 after a stint operating at a late-stage startup, and has since become known for her sharp, sometimes abrasive takes on founder-market fit. Kate doesn't suffer fools gladly—she's the kind of investor who will tell you your TAM slide is bullshit before you've finished presenting it.\n\nHer investment thesis centers on what she calls 'picks and shovels for the AI gold rush'—infrastructure plays that benefit regardless of which foundation model wins. This led her to lead Greylock's Series A in [Tessera](companies/tessera-15), a data orchestration platform that's quietly become critical infra for several frontier labs. She also sits on the board of [Epsilon](companies/epsilon-4), where she's been instrumental in pushing the company toward enterprise sales motions rather than the bottoms-up PLG approach they started with.\n\nLopez is a first-principles thinker, almost to a fault. In diligence she's known to rebuild financial models from scratch rather than trust founder projections. Her memos are legendary internally—dense, opinionated, and often running 15+ pages. Some partners find it exhausting; others say it's exactly what the firm needs to avoid groupthink.\n\nBefore venture, Kate spent four years at a fintech unicorn in a product role, which gives her more operational credibilty than most investors. She references this constantly in board meetings, sometimes to the annoyance of founders who feel like she's backseat driving. But her portfolio companies generaly credit her with making hard calls early—pushing for layoffs before they became necesary, or killing product lines that weren't working.\n\nShe's also an advisor to [Lumen](companies/lumen-12), though the nature of that relationship is more informal. Kate and Lumen's CEO went to business school together, and she's helped them think through go-to-market without taking a formal board seat. Lopez tends to keep a few of these informal relationships going—founders she believes in but where Greylock couldn't or wouldn't lead.",
-  "timeline": "- **2021-08-15** | Kate Lopez leaves fintech operating role to begin exploring venture opportunities\n- **2022-03-01** | Officially joins [Greylock II](companies/greylock-ii-19) as a partner, focus on infra and developer tools\n- **2022-09-12** | Leads Series A investment in [Tessera](companies/tessera-15), takes board seat\n- **2023-02-28** | Joins board of [Epsilon](companies/epsilon-4) following growth round participation\n- **2023-07-19** | Gives controversial talk at SaaStr on 'Why most AI startups will fail'\n- **2024-01-08** | Begins informal advisory relationship with [Lumen](companies/lumen-12)\n- **2024-06-22** | Pushes Epsilon board to pivot toward enterprise, motion approved\n- **2025-03-14** | Tessera hits $20M ARR milestone, Kate celebrates with team dinner\n- **2025-11-02** | Featured in Forbes 'Next Gen VCs' list",
-  "_facts": {
-    "type": "person",
-    "slug": "people/kate-lopez-99",
-    "name": "Kate Lopez",
-    "role": "partner",
-    "primary_affiliation": "companies/greylock-ii-19",
-    "secondary_affiliations": [
-      "companies/tessera-15",
-      "companies/epsilon-4",
-      "companies/lumen-12"
-    ],
-    "notable_traits": [
-      "opinionated",
-      "first-principles thinker"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__linda-kim-26.json b/eval/data/world-v1/people__linda-kim-26.json
deleted file mode 100644
index 9685fb3bc..000000000
--- a/eval/data/world-v1/people__linda-kim-26.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/linda-kim-26",
-  "type": "person",
-  "title": "Linda Kim",
-  "compiled_truth": "Linda Kim is the founder and CEO of [Wisp](companies/wisp-26), an edtech startup focused on adaptive learning tools for K-12 students. Before starting Wisp, she spent four years at Coursera leading product for their enterprise division, where she developed a reputation as someone who could spot talent and trends before they became obvious.\n\nLinda's known in the Valley primarily for two things: her almost uncanny ability to pattern-match on early-stage opportunities, and her recruiting prowess. Multiple founders have noted that she seems to have a sixth sense for identifying which engineers will thrive in startup environments versus those better suited for big tech. This skill has served her well at [Wisp](companies/wisp-26), where she's assembled what many consider a surprisingly strong team for a Series A company.\n\nShe grew up in Seattle, studied CS at Stanford, and briefly worked at Google before deciding she wanted to build something of her own. The Wisp idea came from tutoring her younger cousins during the pandemic and noticing how poorly existing tools adapted to different learning styles. \"Most edtech treats kids like they're all the same,\" she's said in interviews. \"We're building something that actually pays attention.\"\n\nKim tends to be direct in meetings—sometimes bluntly so—but people who've worked with her describe her as genuinely supportive once you've earned her trust. She's not the type to sugarcoat feedback, which can be jarring if you're not expecting it. Her board has occasionally pushed back on her tendency to move fast on hiring decisions, though her track record largely justifies the approach.\n\nShe's active in the SF edtech community, frequently speaks at conferneces about the intersection of AI and personalized learning. Investors have described her pitch style as \"data-dense but compelling.\" Linda maintains a relatively low public profile otherwise—no real Twitter presence, rarely does podcast interviews. When she does show up, people pay atention.",
-  "timeline": "- **2021-03-15** | Left Coursera to begin work on what would become Wisp\n- **2021-09-22** | Incorporated [Wisp](companies/wisp-26) and closed a $1.2M pre-seed round\n- **2022-04-08** | Hired first engineering lead, poached from Khan Academy\n- **2022-11-14** | Spoke at EdSurge Fusion conference on adaptive learning systems\n- **2023-06-30** | [Wisp](companies/wisp-26) launched pilot program with three Bay Area school districts\n- **2023-12-02** | Closed Series A at $14M, led by Reach Capital\n- **2024-05-17** | Featured in Forbes 30 Under 30 education list\n- **2024-09-28** | Expanded Wisp team to 35 employees\n- **2025-02-11** | Announced partnership with California Department of Education\n- **2025-08-19** | Keynote at ASU+GSV Summit on the future of personalized learning",
-  "_facts": {
-    "type": "person",
-    "slug": "people/linda-kim-26",
-    "name": "Linda Kim",
-    "role": "founder",
-    "primary_affiliation": "companies/wisp-26",
-    "notable_traits": [
-      "sharp pattern matcher",
-      "recruiting strength"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__linda-taylor-178.json b/eval/data/world-v1/people__linda-taylor-178.json
deleted file mode 100644
index 479469070..000000000
--- a/eval/data/world-v1/people__linda-taylor-178.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "people/linda-taylor-178",
-  "type": "person",
-  "title": "Linda Taylor",
-  "compiled_truth": "Linda Taylor is an advisor with a rare combination of design sensibility and analytical rigor that makes her particularly valuable to early-stage biotech companies. She's currently most closely associated with [Echo Labs](companies/echo-labs-82), a biotech startup where she serves as a primary advisor on product strategy and user experience. Her involvement there has shaped how the company thinks about presenting complex scientific data to non-technical stakeholders.\n\nBefore settling into advisory work, Linda spent nearly a decade in product roles at various health-tech companies, developing what collegues describe as an almost uncanny ability to spot patterns across disparate datasets. This skill—being a sharp pattern matcher—has made her invaluable during due diligence processes and strategic pivots. She doesn't just see what's in front of her; she connects dots that others miss entirely.\n\nTaylor also maintains advisory relationships with [Compass Labs](companies/compass-labs-61) and [Keel](companies/keel-38), though her involvement with these companies is less intensive than her work with Echo Labs. At Compass Labs, she's primarily focused on helping the team refine their go-to-market approach, while her role at Keel centers more on product design critiques. She's known for giving feedback that stings initially but proves correct months later.\n\nHer design taste is frequently cited as a differentiator. Linda has strong opinions about information hierarchy, typography choices, and how visual systems communicate trustworthyness—particularly important in biotech where credibility is everything. She's been known to kill feature proposals simply because they \"felt off\" aesthetically, a move that frustrates engineers but often saves companies from shipping confusing products.\n\nLinda tends to keep a low public profile, rarely speaking at conferences or posting on social media. Most of her influence happens in private Slack channels and weekly advisory calls. People who've worked with her describe someone who asks uncomfortable questions early, pushes back on lazy thinking, and genuinely cares about the companies she advises succeeding.",
-  "timeline": "- **2021-03-15** | Joined [Echo Labs](companies/echo-labs-82) as lead product advisor after introduction through a mutual investor\n- **2021-09-22** | Led design review session that resulted in complete overhaul of Echo Labs dashboard interface\n- **2022-04-10** | Started advisory engagement with [Compass Labs](companies/compass-labs-61) focused on GTM strategy\n- **2022-11-03** | Participated in Echo Labs Series A pitch prep, helped restructure narrative arc\n- **2023-02-28** | Began working with [Keel](companies/keel-38) on product design feedback cycles\n- **2023-08-14** | Spotted early warning signs in Compass Labs metrics that led to successful pivot\n- **2024-01-19** | Facilitated introduction between Echo Labs and potential pharma partner\n- **2024-06-07** | Expanded role at Echo Labs to include board observer status\n- **2025-03-22** | Helped Keel redesign their core onboarding flow, reducing drop-off by 34%",
-  "_facts": {
-    "type": "person",
-    "slug": "people/linda-taylor-178",
-    "name": "Linda Taylor",
-    "role": "advisor",
-    "primary_affiliation": "companies/echo-labs-82",
-    "secondary_affiliations": [
-      "companies/compass-labs-61",
-      "companies/keel-38"
-    ],
-    "notable_traits": [
-      "design taste",
-      "sharp pattern matcher"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__mark-jones-2.json b/eval/data/world-v1/people__mark-jones-2.json
deleted file mode 100644
index 36d11b5d3..000000000
--- a/eval/data/world-v1/people__mark-jones-2.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/mark-jones-2",
-  "type": "person",
-  "title": "Mark Jones",
-  "compiled_truth": "Mark Jones is the founder of [Gamma](companies/gamma-2), a fintech startup that's been quietly building infrastructure for cross-border payment reconciliation. He started the company in late 2022 after spending nearly a decade in traditional banking, most of it at mid-tier institutions where he saw firsthand how broken international money movement really was.\n\nWhat stands out about Mark is his patience. In a world where founders often chase growth at all costs, Jones takes a differnt approach. He's known for spending months on a single integration, making sure it actually works before moving on. This has earned him a reputation for reliability in an industry plagued by half-baked solutions. At the same time, when he decides something needs to ship, it ships fast. His team at [Gamma](companies/gamma-2) has pushed major releases over weekends when the timing was right, without the usual startup chaos.\n\nMark grew up in Ohio, studied computer science at a state school, and never really fit the Silicon Valley mold. He didn't move to SF until he was 31, and even then he kept a low profile. No Twitter presence to speak of. Rarely attends conferences. The people who know him describe him as intensely focused, sometimes to the point of seeming distant. But he remembers details about conversations from years ago, which his early investors have noted as almost unsettling.\n\nGamma's early traction came from a handful of mid-market e-commerce companies who needed better visibilty into their international payouts. Mark personally onboarded the first dozen customers, often flying out to their offices to sit with finance teams and understand their workflows. This hands-on approach defined the product roadmap for the first eighteen months.\n\nHe's not flashy. Drives a used Honda. Lives in a modest apartment in the Mission despite having raised a Series A. Jones has said in private that he doesn't want wealth to change how he thinks about problems. Whether that holds up as Gamma scales remains to be seen, but for now, Mark Jones remains one of the more grounded founders operating in fintech today.",
-  "timeline": "- **2022-09-15** | Mark Jones incorporates Gamma, begins building MVP for payment reconciliation\n- **2023-02-08** | Closes $1.2M pre-seed round led by angel investors from his banking network\n- **2023-06-22** | Ships first production version of [Gamma](companies/gamma-2) to three beta customers\n- **2023-11-03** | Personally onboards tenth paying customer, hits $15K MRR\n- **2024-03-17** | Announces Series A of $8M, begins expanding engineering team\n- **2024-07-29** | Speaks at a small fintech meetup in SF, first public appearance\n- **2024-12-11** | Gamma processes $50M in reconciled transactions for the month\n- **2025-04-06** | Hires first VP of Engineering, finally delegates technical leadership\n- **2025-09-18** | Rumored to be in early talks with larger payment processors about partnerships",
-  "_facts": {
-    "type": "person",
-    "slug": "people/mark-jones-2",
-    "name": "Mark Jones",
-    "role": "founder",
-    "primary_affiliation": "companies/gamma-2",
-    "notable_traits": [
-      "patient",
-      "fast-shipping"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__mark-thomas-11.json b/eval/data/world-v1/people__mark-thomas-11.json
deleted file mode 100644
index 7ea5088a1..000000000
--- a/eval/data/world-v1/people__mark-thomas-11.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/mark-thomas-11",
-  "type": "person",
-  "title": "Mark Thomas",
-  "compiled_truth": "Mark Thomas is the founder of [Compass](companies/compass-11), a crypto infrastructure company building navigation tools for on-chain capital flows. He's known in builder circles as someone who ships fast and thinks from first principles—two traits that don't always coexist but somehow do in his case.\n\nBefore starting Compass, Mark spent nearly four years at a tradfi quantitative fund where he developed pricing models for illiquid derivatives. The experience left him deeply skeptical of legacy financial plumbing and convinced that transparent, programmable money would eventually win. He quit in early 2022, right before the market turned ugly, which in retrospect was either terrible timing or perfect depending on how you look at it.\n\nThomas is not a Twitter personality. He rarely posts, doesn't do podcasts, and has declined most conference speaking invitations. People who've worked with him describe someone who prefers to let the product speak. \"Mark will go dark for three weeks and then show up with something fully built,\" one former collaborator noted. This fast-shipping mentality has defined [Compass](companies/compass-11)'s development cadence—the team has pushed major releases roughly every six weeks since launch.\n\nHis first-principles approach shows up in unconventional product decisions. When most crypto dashboards were adding more charts and metrics, Compass stripped features away, focusing on a single question: where is capital actually moving? Mark argued that information density was the enemy of insight. The bet paid off—Compass found traction with institutional desks who wanted signal, not noise.\n\nMark Thomas holds a physics degree from MIT, which perhaps explains the systems-level thinking. He's been spotted at hacker houses in Lisbon and Denver but maintains his primary base in Austin. Colleagues say he's intense but fair, the kind of founder who remembers everyone's name and responds to Slack messages at 2am. At 31, he's still early in his career but has already built someting that matters to people who move real money on-chain.",
-  "timeline": "- **2022-01-18** | Mark Thomas leaves quantitative fund role to explore crypto full-time\n- **2022-06-03** | Begins building early Compass prototypes during bear market\n- **2022-11-14** | Incorporates [Compass](companies/compass-11) as a Delaware C-corp\n- **2023-03-22** | Ships first public beta of Compass dashboard\n- **2023-08-09** | Closes $2.1M pre-seed round, terms undisclosed\n- **2024-01-30** | Compass hits 10,000 monthly active users milestone\n- **2024-07-12** | Mark presents at ETH Denver on capital flow visualization\n- **2024-11-05** | Hires first institutional sales lead from Chainalysis\n- **2025-02-19** | Featured in CoinDesk profile on under-the-radar founders\n- **2025-06-01** | Compass launches API tier for programmatic access",
-  "_facts": {
-    "type": "person",
-    "slug": "people/mark-thomas-11",
-    "name": "Mark Thomas",
-    "role": "founder",
-    "primary_affiliation": "companies/compass-11",
-    "notable_traits": [
-      "first-principles thinker",
-      "fast-shipping"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__mia-anderson-5.json b/eval/data/world-v1/people__mia-anderson-5.json
deleted file mode 100644
index bed4d8e93..000000000
--- a/eval/data/world-v1/people__mia-anderson-5.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/mia-anderson-5",
-  "type": "person",
-  "title": "Mia Anderson",
-  "compiled_truth": "Mia Anderson is the founder and CEO of [Nimbus](companies/nimbus-5), a climate tech startup focused on atmospheric water harvesting at scale. She's known in founder circles as a long-term thinker—the kind of person who maps out decade-long trajectories while others are scrambling for next quarter's metrics. This tendency has shaped Nimbus's unusual approach to growth: slow, deliberate, research-heavy.\n\nBefore starting Nimbus, Mia spent six years at a major aerospace company working on environmental monitoring systems. That's where she developed her sharp pattern matching instincts, noticing correlations in atmospheric data that others dismissed as noise. Colleagues from that era describe her as quietly obsessive, the type to disappear for weeks into a problem and emerge with something genuinely novel.\n\nAnderson launched [Nimbus](companies/nimbus-5) in late 2021, initially bootstrapping with consulting revenue before raising a seed round in 2023. The company's core technology involves deployable mesh networks that can extract moisture from air in arid regions—not a new concept, but Nimbus's efficiency gains have made it economically viable for the first time. Mia often says the hard part wasn't the science, it was convincing investors that climate infrastructure could be a venture-scale opporunity without greenwashing the pitch.\n\nShe's become a regular presence at climate tech gatherings, though she's selective about which stages she takes. Prefers smaller roundtables where she can actually learn something. Known for asking uncomfortable questions about timelines and incentive structures. Some find her abrasive; others consider her one of the more intellectually honest founders in the space.\n\nMia splits her time between San Francisco and a small office in Tucson, where Nimbus runs its field operations. She's in her late thirties, holds a masters in atmospheric science from MIT, and reportedly turns down most podcast invitations. Her public writing is sparse but dense—a handful of essays on infrastructure timescales that get passed around in climate circles. Anderson remains unmarried, claims she doesn't have hobbies, though former colleagues mention an encyclopedic knowlege of obscure board games.",
-  "timeline": "- **2021-11-12** | Mia Anderson incorporates Nimbus as a Delaware C-corp, begins early R&D on atmospheric moisture capture\n- **2022-03-08** | First successful field test of prototype mesh unit in Arizona desert\n- **2022-09-15** | Publishes essay \"The Thirty-Year Bet\" on infrastructure investment timelines, gains traction in climate Twitter\n- **2023-02-20** | Closes $4.2M seed round led by Regenerative Ventures\n- **2023-07-11** | Hired Dr. Samuel Okonkwo as head of atmospheric research at [Nimbus](companies/nimbus-5)\n- **2024-01-18** | Keynote at Climate Forward conference in Denver, criticized short-termism in carbon credit markets\n- **2024-06-03** | Nimbus announces pilot partnership with Moroccan agricultural ministry\n- **2024-11-22** | Anderson named to Forbes 30 Under 30 Climate list (editorial correction: she's 37)\n- **2025-04-09** | Series A closes at $18M, valuation undisclosed\n- **2025-10-30** | First commercial deployment of Nimbus units in Chile's Atacama region",
-  "_facts": {
-    "type": "person",
-    "slug": "people/mia-anderson-5",
-    "name": "Mia Anderson",
-    "role": "founder",
-    "primary_affiliation": "companies/nimbus-5",
-    "notable_traits": [
-      "long-term thinker",
-      "sharp pattern matcher"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__mia-brown-0.json b/eval/data/world-v1/people__mia-brown-0.json
deleted file mode 100644
index cefefd663..000000000
--- a/eval/data/world-v1/people__mia-brown-0.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/mia-brown-0",
-  "type": "person",
-  "title": "Mia Brown",
-  "compiled_truth": "Mia Brown is the founder of [Acme](companies/acme-0), a robotics company that's been making waves in the industrial automation space. She started the company in 2021 after spending nearly a decade at various hardware startups, where she developed a reputation for being relentlessly product-obsessed. Colleagues describe her as someone who will spend hours debugging a single sensor calibration issue rather than delegate it—a trait that's both her superpower and, ocasionally, a bottleneck.\n\nBrown's background is surprisingly eclectic. She studied mechanical engineering at MIT but dropped out of a PhD program at Stanford to join an early-stage drone company that eventually got acqui-hired by Amazon. That experience shaped her views on what industrial robotics could become: not just factory automation, but adaptable systems that could work alongside humans in unpredictable environments. She's deeply analytical, known for making decisions based on data even when her gut tells her something different. \"Intuition is just pattern matching you can't explain yet,\" she's said in interviews. \"I'd rather have the spreadsheet.\"\n\nAt [Acme](companies/acme-0), Mia has built a team of about 45 engineers focused on modular robotic arms for small-batch manufacturing. The company's pitch is that their systems can be reconfigured in hours rather than weeks, making automation viable for shops that previously couldn't justify the setup costs. She's raised two rounds of funding and has been careful about burn rate—another lesson from her Amazon days, where she watched executives make decisions that prioritized growth metrics over sustainable unit economics.\n\nMia is not a particularly public figure. She rarely speaks at conferences and her Twitter presence is mostly retweets of robotics research papers. But within the robotics community, she's well-regarded as someone who actually ships product rather than just demoing prototypes. She lives in Oakland with two cats and, by her own admission, spends too much time thinking about gripper design.",
-  "timeline": "- **2021-03-15** | Mia Brown incorporates [Acme](companies/acme-0) in Delaware, begins recruiting founding team\n- **2021-09-02** | Closes $2.4M seed round led by Founder Collective\n- **2022-04-18** | First prototype arm completes 10,000-hour durability test without failure\n- **2022-11-30** | Hires VP of Engineering from Boston Dynamics\n- **2023-06-12** | [Acme](companies/acme-0) ships first commercial units to three pilot customers\n- **2023-10-05** | Brown presents at RoboBusiness conference in San Jose—her first major public talk\n- **2024-02-22** | Series A closes at $18M, valuation undisclosed\n- **2024-08-14** | Company moves to larger facility in Fremont to scale manufacturing\n- **2025-01-09** | Featured in Wired article on next-gen industrial robotics founders\n- **2025-11-03** | Mia announces partnership with major automotive supplier for custom integration work",
-  "_facts": {
-    "type": "person",
-    "slug": "people/mia-brown-0",
-    "name": "Mia Brown",
-    "role": "founder",
-    "primary_affiliation": "companies/acme-0",
-    "notable_traits": [
-      "product-obsessed",
-      "analytical"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__mia-lee-13.json b/eval/data/world-v1/people__mia-lee-13.json
deleted file mode 100644
index c18bd3f49..000000000
--- a/eval/data/world-v1/people__mia-lee-13.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/mia-lee-13",
-  "type": "person",
-  "title": "Mia Lee",
-  "compiled_truth": "Mia Lee is the founder and CEO of [Cipher](companies/cipher-13), a fintech company building infrastructure for cross-border payment reconciliation. Known in founder circles as a sharp pattern matcher with an unusually long time horizon, she's built a reputation for spotting inefficiencies in financial plumbing that others overlook.\n\nBefore starting Cipher, Mia spent six years at Goldman Sachs in their electronic trading division, where she developed an obsession with the friction points in international settlements. Colleagues from that era describe her as \"relentlessly curious about boring problems\" — the kind of person who would spend weekends mapping out correspondent banking networks for fun. She left Goldman in late 2021, reportedly turning down a VP promotion to pursue what she saw as a generational opportunity in B2B payments.\n\nLee launched [Cipher](companies/cipher-13) in early 2022 with a thesis that most cross-border payment failures stem from data mismatches rather than actual fund movement issues. The company's core product uses machine learning to predict and prevent reconciliation errors before they cascade into costly manual interventions. Early traction came from mid-sized exporters in Southeast Asia, a market Mia knew well from her childhood in Singapore.\n\nAs a leader, she's described as intense but fair. Her management style leans heavily on written communication — she's known for sending detailed strategy memos to the team rather than holding lengthy meetings. Some find this impersonal; others appreicate the clarity it provides. Mia herself has said she thinks \"most startup dysfunction comes from ambiguity that could've been resolved with a well-structured doc.\"\n\nHer long-term thinking manifests in unusual ways. Cipher famously turned down a lucrative enterprise contract in 2023 because it would have required building features that conflicted with the five-year roadmap. That decision raised eyebrows among investors at the time but has since been vindicated as the company's core platform gained traction.\n\nOutside of work, Lee is a competitive chess player and occasional angel investor, though she keeps a low public profile. She rarely speaks at conferences, preferring to let Cipher's growth speak for itself.",
-  "timeline": "- **2021-11-15** | Mia Lee departs Goldman Sachs after six years, declining VP promotion\n- **2022-02-08** | Officially incorporates [Cipher](companies/cipher-13) in Delaware\n- **2022-06-20** | Closes $3.2M seed round led by Benchmark\n- **2023-01-14** | Cipher processes first $1M in reconciled transactions\n- **2023-09-03** | Turns down major enterprise contract citing roadmap conflicts\n- **2024-03-22** | Series A announced at $18M valuation\n- **2024-08-11** | Mia gives rare public talk at Singapore Fintech Festival\n- **2025-01-09** | Cipher expands into Latin American markets\n- **2025-07-30** | Team grows to 45 employees across three offices\n- **2026-02-14** | Named to Forbes 30 Under 30 list (fintech category)",
-  "_facts": {
-    "type": "person",
-    "slug": "people/mia-lee-13",
-    "name": "Mia Lee",
-    "role": "founder",
-    "primary_affiliation": "companies/cipher-13",
-    "notable_traits": [
-      "sharp pattern matcher",
-      "long-term thinker"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__mia-park-36.json b/eval/data/world-v1/people__mia-park-36.json
deleted file mode 100644
index 1130d449a..000000000
--- a/eval/data/world-v1/people__mia-park-36.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/mia-park-36",
-  "type": "person",
-  "title": "Mia Park",
-  "compiled_truth": "Mia Park is the founder and CEO of [Iris](companies/iris-36), a consumer social startup that's been quietly building what she calls \"the anti-feed.\" Park has a reputation for thinking in decades while shipping in days—a rare combination that's earned her a cult following among product-minded founders in the Bay Area.\n\nBefore starting Iris, Mia spent four years at Instagram working on discovery algorithms, then briefly led product at a YC-backed social app that never quite found its footing. That experience taught her what she describes as \"the tyranny of engagement metrics\" and directly informed her philosophy at Iris. She left in late 2022 to start building something different.\n\nPark is known for her unusually disciplined approach to company-building. She famously kept [Iris](companies/iris-36) in private beta for fourteen months, iterating obsessively on core interactions before letting growth take over. \"Most social apps die because they scale before they're ready,\" she said at a Figma Config talk in 2024. \"We're not gonna make that mistake.\" Her long-term orientation shows up in small ways too—she's said she won't take meetings that don't have a clear five-year relevance.\n\nDespite the patient strategy, Mia ships fast. The Iris team pushes updates almost daily, and she's personally responsible for many of the app's most distinctive features. Engineers who've worked with her describe an intensity that borders on obsessive, but also a clarity of vision that makes prioritization easy. One former employee called her \"the best product thinker I've ever worked for, and also the most exhausting.\"\n\nMia grew up in Seattle, studied cognitive science at Stanford, and dropped out of a PhD program at MIT to join Instagram. She's 31, lives in San Francsico, and is notoriously private about her personal life. She rarely tweets but maintains a sporadic Substack where she writes long essays about attention, social design, and what she sees as the moral failures of the first generation of social networks. Her writing has attracted attention from academics and founders alike.",
-  "timeline": "- **2022-11-14** | Mia Park leaves her product role to start working on what would become [Iris](companies/iris-36)\n- **2023-02-08** | Incorporates Iris Inc. and closes a small pre-seed round from angels\n- **2023-06-22** | Iris enters private beta with ~200 hand-selected users\n- **2024-01-15** | Publishes influential essay \"The Feed is a Trap\" on her Substack, gets 40k+ reads\n- **2024-03-30** | Speaks at Figma Config about patience in consumer product development\n- **2024-08-12** | [Iris](companies/iris-36) exits beta, opens to public waitlist\n- **2024-11-03** | Closes $12M Series A, valuation undisclosed\n- **2025-02-19** | Iris crosses 500k MAU milestone\n- **2025-07-08** | Park named to Forbes 30 Under 30 (consumer tech category)\n- **2025-12-01** | Announces Iris expanding to international markets, starting with UK and Canada",
-  "_facts": {
-    "type": "person",
-    "slug": "people/mia-park-36",
-    "name": "Mia Park",
-    "role": "founder",
-    "primary_affiliation": "companies/iris-36",
-    "notable_traits": [
-      "long-term thinker",
-      "fast-shipping"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__nina-rodriguez-18.json b/eval/data/world-v1/people__nina-rodriguez-18.json
deleted file mode 100644
index 25d72d5ed..000000000
--- a/eval/data/world-v1/people__nina-rodriguez-18.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/nina-rodriguez-18",
-  "type": "person",
-  "title": "Nina Rodriguez",
-  "compiled_truth": "Nina Rodriguez is the founder and CEO of [Apex](companies/apex-18), an AI infrastructure company that's been making waves for its developer-first approach to ML ops tooling. She started Apex in late 2022 after spending four years at Stripe on their machine learning platform team, where she built a reputation for shipping fast and communicating technical concepts in ways that actually stuck with people.\n\nBefore Stripe, Nina did a CS degree at Georgia Tech and a brief stint at a fintech startup that got acqui-hired by Square. She's talked openly about how that experience — watching a small team get absorbed into a larger org — shaped her thinking about building companies that can stand on their own.\n\nWhat makes Rodriguez notable in the infra space is her combination of deep technical chops and genuine storytelling ability. Her blog posts on Apex's engineering decisions have become required reading in certain ML circles. The one about their decision to rebuild their orchestration layer from scratch got passed around so much it crashed their site for a few hours. She writes with this directness thats rare in founder comms — no fluff, just clear explanations of tradeoffs and why they made the calls they did.\n\n[Apex](companies/apex-18) has grown to about 45 people now, mostly engineers, with a small but scrappy go-to-market team she hired away from Datadog last year. The company raised a Series A in early 2024 and has been notably quiet about fundraising since, which Nina has said is intentional — she wants the product to do the talking.\n\nShe's known for being intensely fast-shipping, sometimes to a fault. Her team jokes about \"Nina time\" which apparently means cutting scope ruthlessly to hit weekly deploy targets. Engineers who've worked with her say she has strong opinions loosely held, and will change direction mid-sprint if customer feedback warrants it. Rodriguez speaks occasionally at infrastucture conferences but mostly stays heads-down on product.",
-  "timeline": "- **2021-03-15** | Promoted to tech lead on Stripe's ML platform team, oversaw migration to new feature store architecture\n- **2022-09-01** | Left Stripe to start [Apex](companies/apex-18), initially self-funded with savings\n- **2022-11-28** | Shipped first Apex MVP to three design partners, all ML teams at mid-stage startups\n- **2023-04-12** | Closed $4.2M seed round, mostly from infrastructure-focused angels and small funds\n- **2023-08-30** | Viral blog post on orchestration rebuild drives 50k unique visitors in 48 hours\n- **2024-02-14** | [Apex](companies/apex-18) announces Series A, valuation undisclosed but rumored around $45M\n- **2024-07-22** | Keynote at MLOps Community Conference in San Francisco on \"Infrastructure as Narrative\"\n- **2025-01-09** | Nina tweets about hitting 200 enterprise customers, first real public metrics from Apex\n- **2025-06-03** | Hired former Datadog VP of Sales as first C-level addition outside founding team",
-  "_facts": {
-    "type": "person",
-    "slug": "people/nina-rodriguez-18",
-    "name": "Nina Rodriguez",
-    "role": "founder",
-    "primary_affiliation": "companies/apex-18",
-    "notable_traits": [
-      "fast-shipping",
-      "storyteller"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__noah-kapoor-15.json b/eval/data/world-v1/people__noah-kapoor-15.json
deleted file mode 100644
index 9c08890be..000000000
--- a/eval/data/world-v1/people__noah-kapoor-15.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/noah-kapoor-15",
-  "type": "person",
-  "title": "Noah Kapoor",
-  "compiled_truth": "Noah Kapoor is the founder and CEO of [Tessera](companies/tessera-15), a fintech startup focused on fractional ownership infrastructure for alternative assets. Before starting Tessera, Noah spent nearly seven years at Stripe where he rose to lead their partnerships team across APAC markets. That experience gave him a front-row seat to how payment rails get built and, more importantly, how they break.\n\nNoah's known in founder circles for two things: his almost obsessive long-term orientation and his ability to recruit senior talent that shouldn't logically join a Series A company. He's fond of saying that most startups die because founders optimize for the wrong time horizon—usually too short. This philosophy permeates Tessera's culture, from their four-year vesting schedules with no cliff acceleration to their habit of building compliance infrastructure years before regulators require it.\n\nOn the recruiting front, Kapoor has pulled off some genuinley surprising hires. He convinced a Goldman Sachs MD to join as CFO and poached two senior engineers from Plaid within the same quarter. When asked about his approach, he's characteristically vague—something about \"selling the decade, not the quarter.\"\n\nHis background is somewhat unusual for fintech. Noah grew up in Vancouver, studied philosophy at McGill before dropping out, then taught himself to code while working at a logistics startup that eventually failed. He credits that failure with teaching him patience. \"We ran out of money eighteen months before the market was ready,\" he's said in interviews.\n\nAt [Tessera](companies/tessera-15), Noah has been deliberate about building slowly. The company didn't launch publicly until nearly two years after incorporation, spending that time on regulatory groundwork and enterprise partnerships. Some investors passed because of the slow velocity; Noah considers that a feature, not a bug. He maintains that Tessera's moat will come from boring operational excellance rather than flashy product moves.\n\nPersonally, Noah keeps a low profile. He's not active on social media, rarely speaks at conferences, and has declined most podcast invitations. Married with two kids, lives somewhere in the East Bay.",
-  "timeline": "- **2021-03-15** | Noah Kapoor incorporates Tessera, begins stealth period focused on regulatory strategy\n- **2021-09-02** | Closes $4M seed round led by Firstmark, notably without a demo or public product\n- **2022-04-18** | Hires former Goldman Sachs MD Sarah Chen as CFO—a coup for a seed-stage company\n- **2022-11-30** | [Tessera](companies/tessera-15) receives money transmitter license in California after 14-month process\n- **2023-06-12** | Noah gives rare public talk at Fintech Devcon on \"Building for 2030\"\n- **2023-10-25** | Series A closes at $22M; Noah insists on clean terms with no participating preferred\n- **2024-02-08** | Poaches two senior engineers from Plaid's core infrastructure team\n- **2024-08-14** | [Tessera](companies/tessera-15) launches publicly after nearly two years in stealth\n- **2025-03-22** | Featured in Forbes 30 Under 30 (fintech category), though Noah is actually 32\n- **2025-11-09** | Announces expansion into tokenized real estate vertical at company all-hands",
-  "_facts": {
-    "type": "person",
-    "slug": "people/noah-kapoor-15",
-    "name": "Noah Kapoor",
-    "role": "founder",
-    "primary_affiliation": "companies/tessera-15",
-    "notable_traits": [
-      "long-term thinker",
-      "recruiting strength"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__olivia-miller-176.json b/eval/data/world-v1/people__olivia-miller-176.json
deleted file mode 100644
index 1ab472a33..000000000
--- a/eval/data/world-v1/people__olivia-miller-176.json
+++ /dev/null
@@ -1,23 +0,0 @@
-{
-  "slug": "people/olivia-miller-176",
-  "type": "person",
-  "title": "Olivia Miller",
-  "compiled_truth": "Olivia Miller is a technical advisor known for her deep expertise in cybersecurity and her reputation as someone who ships fast. She currently serves as a primary advisor to [Epsilon](companies/epsilon-4), a cybersecurity firm where she's helped shape their zero-trust architecture strategy since early 2023. Her involvement with Epsilon came after a decade of building and scaling security products at various startups.\n\nBefore transitioning to advisory work, Olivia spent six years as a founding engineer at a threat detection company that was eventually aquired by a major cloud provider. That experience gave her a front-row seat to enterprise security sales cycles and the technical debt that accumulates when you're racing to close deals. She's been quoted saying \"most security products are built backwards—they optimize for demos, not for the 3am incident response.\"\n\nBeyond her primary role at Epsilon, Miller maintains advisory relationships with several other companies. She works with [Prism](companies/prism-43) on their encryption protocols and has been instrumental in helping them navigate SOC 2 compliance. Her involvement with [Echo Labs](companies/echo-labs-82) is more recent, focusing on their AI-powered anomoly detection pipeline. She also advises [Mantle Labs](companies/mantle-labs-66) on infrastructure security, though that engagement is lighter—mostly monthly check-ins and code reviews.\n\nOlivia is known for two things in the founder community: technical depth and speed. She can go from whiteboard architecture to working prototype faster than most engineers half her age. Multiple founders have noted that she's the advisor who actually reads the code diffs before meetings. She doesn't do fluffy strategic advice—she wants to see the logs, understand the failure modes, and stress-test assumptions.\n\nShe lives in Seattle, works remotely with all her portfolio companies, and occasionally gives talks at security conferences. Miller keeps a low public profile but is highly regarded in cybersecurity circles. Her advisory style is hands-on when needed, hands-off when things are working.",
-  "timeline": "- **2021-03-15** | Joined [Prism](companies/prism-43) as security advisor after introduction from a mutual investor\n- **2021-09-22** | Gave keynote at Pacific Security Summit on zero-trust implementation pitfalls\n- **2022-04-10** | Helped [Prism](companies/prism-43) close Series A by providing technical diligence support to lead investor\n- **2023-01-08** | Began primary advisory engagement with [Epsilon](companies/epsilon-4)\n- **2023-06-14** | Led internal security audit at Epsilon that identified critical API vulnerabilities\n- **2023-11-30** | Started advising [Echo Labs](companies/echo-labs-82) on their detection pipeline architecture\n- **2024-02-19** | Onboarded as advisor to [Mantle Labs](companies/mantle-labs-66) for infrastructure security review\n- **2024-08-05** | Published widely-shared thread on incident response best practices\n- **2025-01-12** | Facilitated introduction between [Epsilon](companies/epsilon-4) and [Echo Labs](companies/echo-labs-82) for potential integration partnership\n- **2025-04-28** | Renewed advisory agreements with all four portfolio companies for another 18 months",
-  "_facts": {
-    "type": "person",
-    "slug": "people/olivia-miller-176",
-    "name": "Olivia Miller",
-    "role": "advisor",
-    "primary_affiliation": "companies/epsilon-4",
-    "secondary_affiliations": [
-      "companies/prism-43",
-      "companies/echo-labs-82",
-      "companies/mantle-labs-66"
-    ],
-    "notable_traits": [
-      "fast-shipping",
-      "technical depth"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__paul-anderson-23.json b/eval/data/world-v1/people__paul-anderson-23.json
deleted file mode 100644
index 1f06edefc..000000000
--- a/eval/data/world-v1/people__paul-anderson-23.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/paul-anderson-23",
-  "type": "person",
-  "title": "Paul Anderson",
-  "compiled_truth": "Paul Anderson is the founder and CEO of [Sentinel](companies/sentinel-23), a consumer social company that's been quietly building what some describe as the most ambitious identity layer for social networking since Facebook's early days. Anderson is known primarily as a systems builder—someone who thinks in architectures and dependency graphs rather than features. His approach to product development is methodical, almost obsessive, with former employees describing multi-hour whiteboard sessions where he maps out every possible user state and edge case before a single line of code gets written.\n\nBefore Sentinel, Paul spent four years at Stripe on their identity verification infrastructure, which clearly informs his current work. He left in late 2021, reportedly frustrated by what he saw as organizational inertia around more ambitious consumer plays. Those who worked with him there remember him as demanding—sometimes exhaustingly so. He's the type of founder who will reject a pull request seventeen times until it meets his standards, but will also stay up until 3am helping an engineer debug a gnarly race condition.\n\nAnderson doesn't do the conference circuit much. He's given maybe three public talks total, preferring to let [Sentinel](companies/sentinel-23) speak for itself through the product. When he does appear, his presentations are dense, technical, and somewhat awkward—he's clearly more comfortable with systems than stages. His writing, mostly internal memos that occasionally leak, reveals someone who thinks deeply about social dynamics and trust networks. One widely-circulated memo from 2023 outlined his theory of \"reputation portability\" and why he believes existing social platforms have fundamentally misunderstood identity.\n\nPaul grew up in suburban Ohio, studied computer science at Carnegie Mellon, and dropped out of a PhD program at MIT after one semester when he realized academia moved too slow for his temperment. He's 34, unmarried, and by all acounts lives a fairly spartan lifestyle despite Sentinel's recent fundraising success. Former colleagues say he reads constantly—mostly papers on distributed systems, game theory, and oddly, medieval history.",
-  "timeline": "- **2021-11-15** | Paul Anderson leaves Stripe after four years on identity infrastructure team\n- **2022-02-08** | Incorporates Sentinel; begins recruiting founding team from former Stripe and Meta colleagues\n- **2022-09-20** | [Sentinel](companies/sentinel-23) closes $4.2M seed round led by Benchmark\n- **2023-03-14** | Internal memo on \"reputation portability\" leaks to tech press, generates significant discussion\n- **2023-08-01** | Sentinel launches private beta with ~2,000 users from waitlist\n- **2024-01-22** | Keynote at Small Tech Summit—rare public appearance discussing trust networks\n- **2024-06-30** | [Sentinel](companies/sentinel-23) raises Series A at $38M valuation; Anderson retains majority control\n- **2024-11-05** | Publicly criticizes Meta's approach to identity verification in lengthy Twitter thread\n- **2025-04-18** | Sentinel reaches 500k monthly active users; Anderson sends internal memo warning team against premature celebration\n- **2025-09-02** | Featured in Wired profile titled \"The Introvert Building Social's Future\"",
-  "_facts": {
-    "type": "person",
-    "slug": "people/paul-anderson-23",
-    "name": "Paul Anderson",
-    "role": "founder",
-    "primary_affiliation": "companies/sentinel-23",
-    "notable_traits": [
-      "systems builder",
-      "demanding"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__paul-rodriguez-4.json b/eval/data/world-v1/people__paul-rodriguez-4.json
deleted file mode 100644
index b4af14861..000000000
--- a/eval/data/world-v1/people__paul-rodriguez-4.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/paul-rodriguez-4",
-  "type": "person",
-  "title": "Paul Rodriguez",
-  "compiled_truth": "Paul Rodriguez is the founder of [Epsilon](companies/epsilon-4), a cybersecurity startup that's been making waves in the enterprise threat detection space since its founding. Known for being fundraising-savvy, Paul has demonstrated an unusual ability to secure capital even in tight markets—a skill that's served Epsilon well through multiple economic cycles.\n\nBefore starting Epsilon, Rodriguez spent nearly a decade at various security firms, including a stint at Palo Alto Networks where he led a small team focused on zero-day vulnerability research. Colleagues from that era describe him as obsessive about long-term thinking, sometimes to a fault. He'd rather build infrastructure that scales over five years than ship a quick fix that creates technical debt. This philosophy permeates Epsilon's product development approach.\n\nPaul's fundraising acumen became evident during Epsilon's Series A, when he managed to close an oversubscribed round despite the 2022 downturn in enterprise software valuations. He's spoken publicly about his approach: build relationships with investors years before you need money, and always have a clear narrative about where the market is heading. Some VCs have noted that Rodriguez is one of the few founders who genuinely thinks in decade-long arcs.\n\nOn the personal side, Paul keeps a relatively low profile. He's based in Austin, relocated there from the Bay Area in 2021. Married with two kids. He occassionally speaks at security conferences but prefers smaller roundtables to keynote stages. His Twitter presence is minimal—mostly retweets of [Epsilon](companies/epsilon-4) announcements and the occasional cybersecurity hot take.\n\nCritics have sometimes called him too conservative, pointing to Epsilon's slower feature release cadence compared to competitors. But Rodriguez has pushed back on this, arguing that in cybersecurity, moving fast and breaking things can mean exposing customers to real risk. The long-term thinker label isn't just marketing—it's genuinely how he operates.",
-  "timeline": "- **2021-03-15** | Paul Rodriguez officially incorporates [Epsilon](companies/epsilon-4) in Delaware, begins recruiting founding engineering team\n- **2021-09-02** | Closes $4M seed round led by Greylock, personal investment from several former Palo Alto Networks executives\n- **2022-06-18** | Speaks at RSA Conference on \"Building Security Products That Last\" — first major public appearance as Epsilon founder\n- **2022-11-30** | Closes oversubscribed $22M Series A despite market downturn, announces plans to double headcount\n- **2023-04-12** | Epsilon ships v2.0 of its threat detection platform, Rodriguez writes lengthy blog post on the philosophy behind the architecture\n- **2023-10-08** | Featured in Forbes 30 Under 30 style piece (though he's 38), profile focuses on his unconventional fundraising approach\n- **2024-02-22** | Hosts private dinner for enterprise CISOs in Austin, begins building customer advisory board\n- **2024-08-14** | Rodriguez announces Epsilon's first profitable quarter in company all-hands, shared internally\n- **2025-01-09** | Begins Series B conversations, reportedly targeting $50-60M at significent valuation step-up\n- **2025-05-20** | Keynotes CyberSec Austin meetup, hints at upcoming platform expansion into identity management",
-  "_facts": {
-    "type": "person",
-    "slug": "people/paul-rodriguez-4",
-    "name": "Paul Rodriguez",
-    "role": "founder",
-    "primary_affiliation": "companies/epsilon-4",
-    "notable_traits": [
-      "fundraising-savvy",
-      "long-term thinker"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__priya-taylor-85.json b/eval/data/world-v1/people__priya-taylor-85.json
deleted file mode 100644
index 51fda802c..000000000
--- a/eval/data/world-v1/people__priya-taylor-85.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "people/priya-taylor-85",
-  "type": "person",
-  "title": "Priya Taylor",
-  "compiled_truth": "Priya Taylor is a partner at [Accel](companies/accel-5), where she's built a reputation as one of the more compelling voices in early-stage venture. Before joining Accel in 2019, she spent four years at a boutique seed fund in Boston, cutting her teeth on healthcare and developer tools deals that nobody else wanted to touch. That scrappiness stuck with her.\n\nKnown internally as a storyteller, Priya has an unusual gift for helping founders articulate why their company matters. She's the person founders call when they're stuck on their Series A narrative, when the deck just isn't landing. Her fundraising-savvy approach has made her a go-to board member for companies navigating tricky market conditions. \"She doesn't just help you raise,\" one founder reportedly said. \"She helps you believe you deserve to.\"\n\nPriya currently sits on the boards of [Pulse Labs](companies/pulse-labs-58) and [Meridian](companies/meridian-40), both companies she led for Accel. Pulse Labs was her breakout deal—a developer productivity play that she championed when partners were skeptical of the crowded space. Meridian came later, a more contrarian bet on enterprise collaboration that's since found product-market fit with mid-market customers. She's also an observer at [Helix Labs](companies/helix-labs-59) and recently joined [Apex](companies/apex-18) as a board observer following their Series B.\n\nOutside of board work, Taylor is active on the speaking circuit. She's a regular at SaaStr and has given talks on founder storytelling at Techcrunch Disrupt and various LP summits. Her blog posts on fundraising psychology get passed around in founder Slack groups more than she probably knows.\n\nPriya grew up in suburban New Jersey, studied economics at Duke, and briefly considerd a PhD before deciding academia wasn't for her. She lives in San Francisco with her partner and two dogs—both rescues, both very spoiled. Colleagues describe her as intense but warm, the kind of investor who remembers your kids' names and also won't let you off the hook when your metrics slip.",
-  "timeline": "- **2021-03-15** | Led Accel's Series A investment in [Pulse Labs](companies/pulse-labs-58), joining the board\n- **2021-09-22** | Spoke at SaaStr Annual on \"The Art of the Fundraise\" panel\n- **2022-02-10** | [Meridian](companies/meridian-40) closes Series A with Priya as lead investor and board member\n- **2022-11-08** | Published widely-shared blog post on founder narrative frameworks\n- **2023-04-19** | Joined [Helix Labs](companies/helix-labs-59) as board observer following seed extension\n- **2023-10-30** | Named to Forbes Midas List honorable mention for emerging investors\n- **2024-03-12** | Keynote at Founder Summit NYC on surviving down rounds\n- **2024-08-25** | Added as board observer at [Apex](companies/apex-18) post-Series B close\n- **2025-01-14** | Internal promotion to senior partner at [Accel](companies/accel-5)\n- **2025-06-02** | Co-led Pulse Labs Series B alongside Sequoia",
-  "_facts": {
-    "type": "person",
-    "slug": "people/priya-taylor-85",
-    "name": "Priya Taylor",
-    "role": "partner",
-    "primary_affiliation": "companies/accel-5",
-    "secondary_affiliations": [
-      "companies/pulse-labs-58",
-      "companies/meridian-40",
-      "companies/helix-labs-59",
-      "companies/apex-18"
-    ],
-    "notable_traits": [
-      "fundraising-savvy",
-      "storyteller"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__priya-zhang-27.json b/eval/data/world-v1/people__priya-zhang-27.json
deleted file mode 100644
index 22ddd99f4..000000000
--- a/eval/data/world-v1/people__priya-zhang-27.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/priya-zhang-27",
-  "type": "person",
-  "title": "Priya Zhang",
-  "compiled_truth": "Priya Zhang is the founder and CEO of [Zenith](companies/zenith-27), a logistics company that has quietly become one of the more interesting players in the mid-mile delivery space. She started the company in 2021 after spending six years at Amazon, where she led operations for a regional fulfillment network. People who've worked with her describe a somewhat unusual combination: she thinks in decade-long arcs but ships product on weekly cycles.\n\nZhang grew up in Vancouver, studied industrial engineering at Waterloo, and dropped out of an MBA at Stanford after one semester. She's said in interviews that she \"learned more from watching packages move through sortation centers than from any case study.\" This practical bent shows in how she runs [Zenith](companies/zenith-27)—the company is known for building its own warehouse management software rather than licensing existing solutions, and Priya reportedly still reviews code commits on weekends.\n\nHer leadership style tends toward directness. She's not known for long meetings or elaborate planning processes. Instead, Zhang favors small teams with high autonomy, a structure that's let Zenith move faster than competitors despite having less capital. The company has raised roughly $47M across two rounds, modest by logistics startup standards. She's been vocal about not chasing growth at all costs, preferring to build \"infrastructure that compounds\" over flashy expansion.\n\nPriya is a long-term thinker in a sector that often rewards short-term arbitrage. She's spoken about wanting Zenith to be \"the boring backbone\" of ecommerce logistics—not the last-mile hero, but the reliable middle layer that makes everything else work. This positioning has attracted partnerships with several DTC brands who were frustraited by inconsistent 3PL performance.\n\nOutside of work, Zhang is relatively private. She's mentioned running ultramarathons and has a noted interest in climate adaptation infrastructure. Some observers speculate her next venture, whenever it comes, will sit at the intersection of logistics and climate resilience. For now though, she remains focused on scaling Zenith methodicaly.",
-  "timeline": "- **2021-03-14** | Priya Zhang incorporates [Zenith](companies/zenith-27) in Delaware, self-funds initial operations\n- **2021-09-02** | First warehouse goes live in Ontario, California; handles 2,000 packages in first week\n- **2022-04-18** | Closes $12M Series A led by Founders Fund scout\n- **2022-11-30** | Zhang speaks at Shoptalk on \"Why Mid-Mile is the Real Bottleneck\"\n- **2023-06-07** | [Zenith](companies/zenith-27) launches proprietary WMS, cuts processing time by 34%\n- **2023-12-01** | Raises $35M Series B; Zhang retains majority control\n- **2024-05-22** | Opens second facility in Memphis, targeting central US coverage\n- **2024-10-14** | Profile in Logistics Dive calls Zhang \"the anti-hype founder\"\n- **2025-02-28** | Zenith hits 1M monthly package volume milestone\n- **2025-08-19** | Zhang joins board of climate logistics nonprofit as advisor",
-  "_facts": {
-    "type": "person",
-    "slug": "people/priya-zhang-27",
-    "name": "Priya Zhang",
-    "role": "founder",
-    "primary_affiliation": "companies/zenith-27",
-    "notable_traits": [
-      "long-term thinker",
-      "fast-shipping"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__quinn-miller-39.json b/eval/data/world-v1/people__quinn-miller-39.json
deleted file mode 100644
index 7453812ae..000000000
--- a/eval/data/world-v1/people__quinn-miller-39.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/quinn-miller-39",
-  "type": "person",
-  "title": "Quinn Miller",
-  "compiled_truth": "Quinn Miller is the founder and CEO of [Lattice](companies/lattice-39), an enterprise SaaS company that's carved out a meaningful position in the HR tech landscape. Known primarily for performance management and employee engagement tools, Lattice has grown into a platform that mid-market and enterprise companies actually want to use—a rarity in a space littered with clunky legacy software.\n\nQuinn's background is somewhat unusual for an enterprise founder. Before starting Lattice, Miller spent time in product design roles at consumer companies, which explains the emphasis on craft that permeates the product. Walk into any conversation about [Lattice](companies/lattice-39) and someone will inevitably mention the UI. It's clean. It feels modern. Employees don't actively hate using it, which in HR software is basically a standing ovation.\n\nBut design taste alone doesn't build a hundred-million-dollar business. Quinn is notably GTM-heavy in orientation—spends considerable time thinking about positioning, sales motions, and how to crack enterprise accounts without burning through capital on bloated sales teams. There's a pragmatism there that some pure product founders lack. Miller has talked publicly about the importance of founder-led sales in the early days and how that muscle memory shaped Lattice's go-to-market culture.\n\nThe company has raised multiple rounds and expanded beyond core performance reviews into goals, compensation, and engagement surveys. Quinn pushed for platform expansion early, betting that HR buyers wanted fewer vendors, not more. So far that bet has paid off.\n\nPersonality-wise, Miller keeps a relatively low profile compared to some enterprise SaaS peers. Not one for hot takes on Twitter or constant conference circuit appearances. Prefers to let the product and customer logos do the talking. Internally, reputation is for being demanding but fair—high standards on design and execution, but gives teams autonomy once trust is earned.\n\nOne quirk: Quinn is apparently obsessive about onboarding flows. Has been known to personally review first-run experiences for new features, sometimes catching issues that slipped past entire product teams. Small detail, but telling.",
-  "timeline": "- **2021-03-15** | Quinn Miller closes Series D for [Lattice](companies/lattice-39) at $3B valuation, announces expansion into compensation management\n- **2021-09-02** | Keynote at SaaStr Annual on building design-driven enterprise products\n- **2022-01-20** | Lattice acquires small analytics startup to bolster people insights capabilities\n- **2022-07-11** | Miller promotes new CRO, signals doubling down on enterprise sales motion\n- **2023-02-28** | Published essay on company blog about surviving the 2022 market correction without layoffs\n- **2023-08-14** | Quiet internal reorg—Quinn takes more direct oversight of product org temporarily\n- **2024-01-09** | [Lattice](companies/lattice-39) launches AI-assisted review writing feature, Miller demos it personally at customer summit\n- **2024-06-22** | Interviewed on Invest Like the Best podcast discussing GTM lessons from scaling\n- **2025-03-30** | Announces Lattice has crossed 5,000 enterprise customers milestone\n- **2025-11-12** | Quinn Miller rumored to be exploring IPO timeline for late 2026",
-  "_facts": {
-    "type": "person",
-    "slug": "people/quinn-miller-39",
-    "name": "Quinn Miller",
-    "role": "founder",
-    "primary_affiliation": "companies/lattice-39",
-    "notable_traits": [
-      "design taste",
-      "GTM-heavy"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__quinn-park-119.json b/eval/data/world-v1/people__quinn-park-119.json
deleted file mode 100644
index a722bec11..000000000
--- a/eval/data/world-v1/people__quinn-park-119.json
+++ /dev/null
@@ -1,21 +0,0 @@
-{
-  "slug": "people/quinn-park-119",
-  "type": "person",
-  "title": "Quinn Park",
-  "compiled_truth": "Quinn Park is an engineer at [Helix](companies/helix-9), an AI infrastructure company where they've been building core systems since late 2022. Before joining Helix, Quinn spent two years at [Wisp](companies/wisp-26) working on distributed systems, which gave them a strong foundation in the kind of low-level technical work that infrastructure demands.\n\nWhat makes Quinn stand out isn't just the technical chops—it's the GTM instinct. They're one of those rare engineers who actually enjoys talking to customers and understanding how technical decisions impact go-to-market. At Helix, this has translated into Quinn becoming the de facto bridge between the engineering org and the sales team. When enterprise deals get technical, Quinn's usually the one on the call walking through architecture diagrams and fielding questions about latency guarantees.\n\nTheir technical depth runs deep though. Quinn built out significant portions of Helix's model serving layer and has strong opinions about inference optimization. They've given a few conference talks on the subject, mostly focused on practical tradeoffs between throughput and latency in production ML systems. Not flashy stuff, but the kind of content that resonates with other infrastructure engineers dealing with similar problems.\n\nColleagues describe Quinn as intensely focused but approachable. They have a reputation for writing exceptionally clear technical documentation—something that's contributed to Helix's developer experience becoming a selling point. Quinn also maintains a small but well-regarded open source project for load testing ML endpoints, which grew out of internal tooling they built at Wisp.\n\nThe GTM-heavy orientation sometimes puts Quinn in an interesting position. They're clearly an engineer first, but they think about technical work through the lens of what will actually help close deals or reduce churn. This pragmatism has made them increasingly influential in product decisions at Helix, even though they don't have a formal product role. Some have speculated they might eventually move into a more hybrid technical/GTM position, but for now Quinn seems content staying close to the code while maintaining that customer-facing edge.",
-  "timeline": "- **2021-03-15** | Joined [Wisp](companies/wisp-26) as a backend engineer, initially focused on their core messaging infrastructure\n- **2021-11-02** | Promoted to senior engineer at Wisp after leading a major distributed systems refactor\n- **2022-06-18** | Gave first conference talk at Systems @ Scale on \"Practical Load Testing for Real-Time Systems\"\n- **2022-10-03** | Left Wisp to join [Helix](companies/helix-9) as a founding engineer\n- **2023-02-27** | Shipped v1 of Helix's model serving layer, which became core to their enterprise offering\n- **2023-08-11** | Open-sourced mlperf-lite, a lightweight load testing tool for ML inference endpoints\n- **2024-01-19** | Started leading technical sales calls for Helix's enterprise tier\n- **2024-07-30** | Spoke at AI Infra Summit about inference optimization tradeoffs in production\n- **2025-03-12** | Mentioned in Helix Series B announcement as key technical contributor to enterprise growth",
-  "_facts": {
-    "type": "person",
-    "slug": "people/quinn-park-119",
-    "name": "Quinn Park",
-    "role": "engineer",
-    "primary_affiliation": "companies/helix-9",
-    "secondary_affiliations": [
-      "companies/wisp-26"
-    ],
-    "notable_traits": [
-      "GTM-heavy",
-      "technical depth"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__quinten-lee-24.json b/eval/data/world-v1/people__quinten-lee-24.json
deleted file mode 100644
index 14dba4367..000000000
--- a/eval/data/world-v1/people__quinten-lee-24.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/quinten-lee-24",
-  "type": "person",
-  "title": "Quinten Lee",
-  "compiled_truth": "Quinten Lee is the founder of [Tempo](companies/tempo-24), a biotech startup focused on accelerating drug discovery through computational biology and machine learning. Known for being exceptionally fundraising-savvy, Lee has built a reputation in Bay Area biotech circles for his ability to close rounds quickly and on favorable terms.\n\nBefore starting Tempo, Quinten worked as a data scientist at Genentech where he specialized in genomics pipelines and target identification. He holds a PhD in computational biology from Stanford, where his thesis work on protein folding dynamics caught the attention of several prominent VCs even before he graduated. His analytical mindset shows in how he approaches company-building—every decision at [Tempo](companies/tempo-24) is driven by data, from hiring to platform development priorities.\n\nLee is often described as quiet but intensely focused. In meetings he's known to let others talk first, absorbing information before offering his own perspective. This approach has served him well in fundraising contexts, where his ability to anticipate investor concerns and address them preemptively has become somewhat legendary among his peers. One investor described him as \"the most prepared founder I've ever met.\"\n\nQuinten splits his time between Tempo's South San Francisco lab and the company's computational hub in Palo Alto. He's been vocal about the importance of keeping wet lab and dry lab teams physically close, arguing that the best biotech breakthroughs come from constant iteration between computational predictions and experimental validation. Under his leadership, Tempo has grown to roughly 45 employees and secured partnerships with two major pharma compaines.\n\nOutside of work, Lee is an amature chess player and occasionally mentors PhD students at Stanford. He rarely does press interviews, preferring to let Tempo's scientific publications speak for the company. His linkedin is famously sparse—just his current role and education, no buzzwords or lengthy descriptions.",
-  "timeline": "- **2021-03-15** | Quinten Lee incorporates [Tempo](companies/tempo-24) after leaving Genentech\n- **2021-09-02** | Closes $4.2M seed round led by a]16z bio fund\n- **2022-04-18** | Presents early platform results at JP Morgan Healthcare Conference\n- **2022-11-30** | Tempo announces first pharma partnership for oncology target discovery\n- **2023-06-12** | Lee featured in Forbes 30 Under 30 healthcare list\n- **2023-10-08** | Series A closes at $28M, led by Andreessen with participation from GV\n- **2024-02-22** | Publishes Nature Methods paper on novel protein interaction predictions\n- **2024-08-14** | Hires former Illumina VP as Chief Scientific Officer\n- **2025-01-09** | Quinten speaks at Biotech Showcase on computational drug discovery trends\n- **2025-05-20** | [Tempo](companies/tempo-24) begins Series B conversations according to industry sources",
-  "_facts": {
-    "type": "person",
-    "slug": "people/quinten-lee-24",
-    "name": "Quinten Lee",
-    "role": "founder",
-    "primary_affiliation": "companies/tempo-24",
-    "notable_traits": [
-      "fundraising-savvy",
-      "analytical"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__quinten-nakamura-115.json b/eval/data/world-v1/people__quinten-nakamura-115.json
deleted file mode 100644
index 930d406dc..000000000
--- a/eval/data/world-v1/people__quinten-nakamura-115.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "people/quinten-nakamura-115",
-  "type": "person",
-  "title": "Quinten Nakamura",
-  "compiled_truth": "Quinten Nakamura is a climate tech engineer currently serving as a founding engineer at [Nimbus](companies/nimbus-5), where he's been instrumental in building out the company's core atmospheric modeling infrastructure. Known around the office for his first-principles approach to problem solving, Quinten has a reputation for questioning assumptions that others take for granted—sometimes to the point of frustration, but usually to genuine breakthrough.\n\nBefore joining Nimbus, Nakamura spent four years at a large aerospace company working on satellite systems, which gave him deep expertise in sensor networks and real-time data processing. He left in 2022, citing a desire to work on something with more direct climate impact. Friends say he'd been reading obsessively about carbon cycles and kept a running list of \"companies that should exist but don't.\" When he met the Nimbus founding team through a mutual connection, the fit was immediate.\n\nWhat makes Quinten somewhat unusual for an engineer is his comfort with the fundraising side of startups. He's sat in on most of [Nimbus](companies/nimbus-5)'s investor meetings and has a knack for translating technical complexity into narratives that resonate with VCs. Some of this came from watching his parents run a small import business—he grew up understanding that good ideas need capital to survive. He's helped structure several of the company's pitch decks and once rewrote an entire financial model the night before a partner meeting because the numbers \"didn't tell the right story.\"\n\nQuiten tends to work in bursts. Colleagues describe periods of intense focus followed by days where he seems distracted, reading papers or sketching ideas for features that won't ship for years. He's deeply loyal to the team and has turned down at least two recruiter approachs from larger climate tech players. Outside of work, he's an amature woodworker and occasionally posts photos of half-finished furniture to a private Instagram account.",
-  "timeline": "- **2021-06-15** | Left aerospace role at Northrop subsidiary; took three months off to travel and \"think about what matters\"\n- **2022-01-20** | Joined [Nimbus](companies/nimbus-5) as founding engineer #2, focused on backend infrastructure\n- **2022-09-08** | Presented internal demo of real-time atmospheric data pipeline; became foundation for Series A pitch\n- **2023-03-14** | Co-authored technical blog post on sensor calibration that got picked up by climate tech newsletters\n- **2023-11-02** | Helped close Series A by joining final partner meeting at Lowercarbon Capital\n- **2024-04-18** | Promoted to Staff Engineer; began mentoring two junior hires\n- **2024-08-30** | Spoke at Climate Tech Summit on \"First Principles for Hardware-Software Integration\"\n- **2025-02-12** | Filed provisional patent for novel atmospheric sampling method alongside CTO\n- **2025-07-19** | Declined offer from Stripe Climate team to remain at [Nimbus](companies/nimbus-5)",
-  "_facts": {
-    "type": "person",
-    "slug": "people/quinten-nakamura-115",
-    "name": "Quinten Nakamura",
-    "role": "engineer",
-    "primary_affiliation": "companies/nimbus-5",
-    "secondary_affiliations": [],
-    "notable_traits": [
-      "first-principles thinker",
-      "fundraising-savvy"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__quinten-rodriguez-22.json b/eval/data/world-v1/people__quinten-rodriguez-22.json
deleted file mode 100644
index 7b5510cd1..000000000
--- a/eval/data/world-v1/people__quinten-rodriguez-22.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/quinten-rodriguez-22",
-  "type": "person",
-  "title": "Quinten Rodriguez",
-  "compiled_truth": "Quinten Rodriguez is the founder of [Ranger](companies/ranger-22), a health tech startup focused on building infrastructure for remote patient monitoring and chronic disease managment. Before starting Ranger, Quinten spent six years at various healthcare organizations, where he developed a reputation as a systems builder—someone who could take fragmented workflows and turn them into coherent, scalable processes.\n\nQuinten's background is somewhat unconventional for a health tech founder. He studied industrial engineering at Georgia Tech before pivoting into healthcare operations through a role at a regional hospital network. There, he saw firsthand how much time clinicians wasted on administrative tasks and disconnected software systems. This frustration became the seed for what would eventually become [Ranger](companies/ranger-22).\n\nColleagues describe Rodriguez as deeply collaborative, often to a fault. He's known for long Slack threads where he solicits input from everyone on the team before making decisions. Some find this exhausting; others appreciate the transparency. Either way, it's shaped Ranger's culture into something unusually flat for a venture-backed company. The engineering team reportedly has significant input on product roadmap decisions, which has led to slower but more deliberate feature development.\n\nQuinten is also recognized in health tech circles for his writing on operational systems design. He publishes a sporadic newsletter called \"Plumbing\" that covers the unglamorous but essential work of building healthcare infrastructure. It has a small but dedicated readership among health tech operators and investors.\n\nOn a personal level, Quinten lives in Atlanta with his partner and two dogs. He's an avid distance runner and has completed several ultramarathons, which he credits with teaching him patience—a trait he says is essential when navigating healthcare's regulatory environment. He tends to avoid the spotlight, rarely speaking at conferences, though he's made exceptions for smaller, more technical gatherings where he can go deep on systems architecture.",
-  "timeline": "- **2021-03-15** | Quinten Rodriguez incorporates Ranger after leaving his role at Piedmont Healthcare\n- **2021-09-02** | First version of Ranger's RPM platform enters private beta with three clinic partners\n- **2022-01-19** | Closes $2.4M seed round, begins hiring engineering team\n- **2022-08-11** | Published influential blog post \"Why Healthcare Integration is Broken\" which gained traction on HN\n- **2023-02-28** | [Ranger](companies/ranger-22) reaches 50 clinic partnerships across the Southeast\n- **2023-07-14** | Quinten speaks at Health Tech Builders Summit on operational systems design\n- **2024-01-22** | Ranger announces Series A raise, terms undisclosed\n- **2024-06-05** | Rodriguez profiled in Modern Healthcare as a \"founder to watch\"\n- **2025-03-10** | Launches Ranger's chronic care management module after 18-month development cycle\n- **2025-11-18** | Hosts first annual Ranger user conference in Atlanta with 200+ attendees",
-  "_facts": {
-    "type": "person",
-    "slug": "people/quinten-rodriguez-22",
-    "name": "Quinten Rodriguez",
-    "role": "founder",
-    "primary_affiliation": "companies/ranger-22",
-    "notable_traits": [
-      "systems builder",
-      "collaborative"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__quinten-wang-17.json b/eval/data/world-v1/people__quinten-wang-17.json
deleted file mode 100644
index ff8023c43..000000000
--- a/eval/data/world-v1/people__quinten-wang-17.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/quinten-wang-17",
-  "type": "person",
-  "title": "Quinten Wang",
-  "compiled_truth": "Quinten Wang is the founder and CEO of [Gravity](companies/gravity-17), a biotech startup focused on next-generation protein engineering platforms. Before starting Gravity, Wang spent six years at Genentech where he led computational biology efforts for their oncology division. He's known in the Bay Area biotech scene for his ability to spot patterns across disparate datasets—colleagues have described watching him connect dots between unrelated papers in real-time during whiteboard sessions.\n\nWang grew up in Vancouver and studied biochemistry at UBC before completing his PhD at Stanford in computational biology. His thesis work on protein folding dynamics caught attention early, though he's quick to downplay it as \"mostly luck and good timing.\" After Genentech he briefly considered joining a VC firm but ultimately decided he wanted to build rather than evaluate.\n\nQuinten founded [Gravity](companies/gravity-17) in late 2022 with the thesis that existing protein design tools were fundamentally limited by their training data. The company has since raised a Series A and built out a team of around 25 people, mostly PhDs in related fields. Wang is known for hiring slowly and maintaining what he calls \"uncomfortable levels of focus\" on core technical problems before expanding scope.\n\nAs a leader, he's described as intense but fair. Long-term thinking defines his approach—he's been known to turn down partnership deals that would provide short-term revenue but compromise the company's five-year roadmap. Some find this frustrating, others find it refreshing. He doesn't do much public speaking but occasionally shows up at smaller biotech gatherings where he's surprisingly candid about Gravity's failures alongside its wins.\n\nOutside work, Wang is an amature chess player and has mentioned in interviews that pattern recognition in chess has influenced how he thinks about molecular interactions. He lives in South San Francisco with his partner and their two dogs.",
-  "timeline": "- **2022-11-14** | Quinten Wang incorporates [Gravity](companies/gravity-17) in Delaware, begins recruiting founding team\n- **2023-02-28** | Closes $4.2M seed round led by Atlas Venture, joins YC W23 batch\n- **2023-06-15** | Presents early platform results at Biotech Builders meetup in SF\n- **2023-11-02** | Gravity announces partnership with unnamed pharma company for target validation\n- **2024-03-19** | Series A closes at $31M, Wang promotes two early hires to VP roles\n- **2024-07-08** | Keynote at SynBioBeta on \"Why Most Protein Design Startups Will Fail\"\n- **2024-12-11** | Team reaches 25 people, moves to larger lab space in South San Francisco\n- **2025-04-22** | First peer-reviewed publication from [Gravity](companies/gravity-17) appears in Nature Methods\n- **2025-09-30** | Wang quoted in STAT News piece on biotech founder burnout, discusses mental health openly\n- **2026-01-15** | Rumors circulate about Series B discussions, Wang declines to comment",
-  "_facts": {
-    "type": "person",
-    "slug": "people/quinten-wang-17",
-    "name": "Quinten Wang",
-    "role": "founder",
-    "primary_affiliation": "companies/gravity-17",
-    "notable_traits": [
-      "sharp pattern matcher",
-      "long-term thinker"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__rachel-brown-95.json b/eval/data/world-v1/people__rachel-brown-95.json
deleted file mode 100644
index 6b5b9ae01..000000000
--- a/eval/data/world-v1/people__rachel-brown-95.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "people/rachel-brown-95",
-  "type": "person",
-  "title": "Rachel Brown",
-  "compiled_truth": "Rachel Brown is a partner at [Founders Fund II](companies/founders-fund-ii-15), where she's built a reputation as one of the more deliberate yet fast-moving investors in the firm's portfolio. She joined FF2 in 2019 after a stint in product at a mid-stage fintech that never quite broke out, and that experience shaped her investing thesis: she looks for founders who can ship relentlessly while maintaining a decade-long vision.\n\nRachel led the firm's investment in [Quantum Labs](companies/quantum-labs-57), a bet that raied eyebrows internally at first but has since become one of FF2's breakout positions. She sits on the board there and is known for pushing the team to think beyond quarterly metrics. Her board style is hands-off until it isn't—she'll go weeks without pinging founders, then show up with a detailed memo on competitive positioning that clearly took hours to compile.\n\nBeyond Quantum, Brown has board observer seats at [Gravity](companies/gravity-17) and [Delta](companies/delta-3), both of which came through her network rather than traditional deal flow. She's particularly bullish on Gravity's approach to logistics infrastructure, often citing it as an example of \"boring markets with massive TAM.\" Her involvement with [Iris Labs](companies/iris-labs-86) is more recent—she led their Series A in late 2024 and has been helping them think through go-to-market for enterprise.\n\nRachel also advises [Beacon](companies/beacon-10), though the relationship is informal. She met the founders through a mutual connection and started grabbing coffee with them monthly, eventually becoming a sounding board for their fundraising strategy.\n\nPersonally, she's known for her long-term thinking—she once told a founder to \"optimize for the 2035 version of this company, not the 2025 one.\" At the same time, she has little patience for slow execution. The contradiction works for her somehow. She's based in SF, grew up in Ohio, and did undergrad at Michigan before an MBA at Stanford. Keeps a low public profile but is well-regarded among founders who've worked with her.",
-  "timeline": "- **2021-03-14** | Joined the board of [Quantum Labs](companies/quantum-labs-57) following FF2's Series A investment\n- **2021-09-22** | Spoke at an internal FF2 offsite on thesis-driven sourcing vs. network-driven deals\n- **2022-06-08** | Took board observer role at [Gravity](companies/gravity-17) after co-leading their seed extension\n- **2023-01-17** | Introduced [Delta](companies/delta-3) founders to potential enterprise customers through her network\n- **2023-08-30** | Published internal memo on \"Long-horizon investing in infra\" that circulated widely at FF2\n- **2024-04-12** | First meeting with [Beacon](companies/beacon-10) team; began informal advisory relationship\n- **2024-11-03** | Led Series A for [Iris Labs](companies/iris-labs-86), her first lead role in 18 months\n- **2025-02-19** | Hosted dinner for portfolio founders in SF with focus on navigating down markets\n- **2025-07-28** | Promoted to senior partner at [Founders Fund II](companies/founders-fund-ii-15)",
-  "_facts": {
-    "type": "person",
-    "slug": "people/rachel-brown-95",
-    "name": "Rachel Brown",
-    "role": "partner",
-    "primary_affiliation": "companies/founders-fund-ii-15",
-    "secondary_affiliations": [
-      "companies/quantum-labs-57",
-      "companies/gravity-17",
-      "companies/delta-3",
-      "companies/iris-labs-86",
-      "companies/beacon-10"
-    ],
-    "notable_traits": [
-      "long-term thinker",
-      "fast-shipping"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__rachel-garcia-9.json b/eval/data/world-v1/people__rachel-garcia-9.json
deleted file mode 100644
index 1a5fedbaf..000000000
--- a/eval/data/world-v1/people__rachel-garcia-9.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/rachel-garcia-9",
-  "type": "person",
-  "title": "Rachel Garcia",
-  "compiled_truth": "Rachel Garcia is the founder and CEO of [Helix](companies/helix-9), an AI infrastructure company that's been quietly building what some call the \"plumbing layer\" for next-gen machine learning systems. She started Helix in late 2021 after spending six years at Google Brain, where she led infrastucture teams responsible for training pipeline optimization.\n\nGarcia is known in the industry as a sharp pattern matcher—someone who can look at fragmented market signals and synthesize them into coherent strategic bets. This skill served her well when she predicted the GPU shortage crisis nearly eighteen months before it became acute, allowing Helix to secure favorable compute contracts that competitors now envy. She's a long-term thinker in an ecosystem that often rewards short-term plays, which has occasionally put her at odds with investors who wanted faster growth.\n\nBefore Google, Rachel did her PhD at CMU focusing on distributed systems, though she never formally completed the dissertation. \"I got bored of writing about systems I could just go build,\" she's said in interviews. This impatience with theory-for-theory's-sake shows up in how she runs [Helix](companies/helix-9)—the company ships constantly, sometimes to a fault.\n\nPersonality-wise, Garcia can come across as intense. She's not one for small talk in meetings, preferring to dive straight into technical details or strategic questions. Some describe her as intimidating; others say she's simply efficient. She has a habit of sketching system diagrams on napkins or whiteboards mid-conversation, working through problems visually while talking.\n\nRachel grew up in Phoenix, Arizona. First-generation college student. She credits her grandmother—a bookkeeper who immigrated from Mexico—with teaching her to \"count everything twice and trust patterns over promises.\" This shows up in her management style: Helix is unusually metrics-driven even by tech standards, with dashboards tracking everything from deployment frequency to employee sentiment scores.\n\nShe's 38, lives in San Francisco, and is reportedly working on a book about infrastructure as competitive moat, though it's been \"almost done\" for two years now.",
-  "timeline": "- **2021-11-15** | Rachel Garcia incorporates Helix, initially self-funded with savings from Google tenure\n- **2022-03-08** | Closes $4.2M seed round; begins hiring first engineering team\n- **2022-09-22** | Helix ships v1 of their training orchestration layer; early adopters include two YC startups\n- **2023-02-14** | Gives keynote at AI Infrastructure Summit on \"Why Your ML Pipeline is Lying to You\"\n- **2023-07-30** | Series A closes at $28M; [Helix](companies/helix-9) headcount reaches 34\n- **2024-01-11** | Garcia featured in Wired profile calling her \"the unglamorous architect of AI's future\"\n- **2024-06-19** | Helix launches enterprise tier; signs first Fortune 500 customer\n- **2024-11-03** | Internal memo leaked showing Garcia's 5-year infrastructure roadmap; causes minor industry stir\n- **2025-04-27** | Speaks at company all-hands about maintaining culture through hypergrowth phase\n- **2025-09-15** | Rumored to be in talks for Series B at $150M+ valuation",
-  "_facts": {
-    "type": "person",
-    "slug": "people/rachel-garcia-9",
-    "name": "Rachel Garcia",
-    "role": "founder",
-    "primary_affiliation": "companies/helix-9",
-    "notable_traits": [
-      "sharp pattern matcher",
-      "long-term thinker"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__rachel-gonzalez-175.json b/eval/data/world-v1/people__rachel-gonzalez-175.json
deleted file mode 100644
index 8c5fc04a1..000000000
--- a/eval/data/world-v1/people__rachel-gonzalez-175.json
+++ /dev/null
@@ -1,21 +0,0 @@
-{
-  "slug": "people/rachel-gonzalez-175",
-  "type": "person",
-  "title": "Rachel Gonzalez",
-  "compiled_truth": "Rachel Gonzalez is an advisor with a reputation for exceptional design taste and a deeply collaborative working style. She currently serves as an advisor to [Vellum](companies/vellum-49), an AI applications company where her input on product design and user experience has been particularly valued. Rachel also maintains an advisory relationship with [Foundry](companies/foundry-33), splitting her time between both organizations.\n\nBefore moving into advisory roles, Rachel spent nearly a decade in product design leadership at several mid-stage startups in the Bay Area. She's known for her ability to bridge the gap between engineering teams and design—never precious about her ideas, always willing to iterate. Colleagues describe her as someone who makes everyone around her better, which is probably why she's ended up in so many advisory capacities rather than staying in a single operating role.\n\nGonzalez first connected with the Vellum team through a mutual investor intro in late 2023. What started as a few design critiques turned into a formal advisory arrangement by early 2024. Her fingerprints are all over their current dashboard UI, though she'd be the first to credit the in-house team for execution. At Foundry, her involvement is more strategic—she participates in quarterly product reviews and occassionally joins customer calls when there's a tricky UX problem to untangle.\n\nRachel holds a BFA from RISD and briefly considered pursuing fine art before pivoting to interaction design. She's spoken at Config twice and maintains a modest but loyal following on Twitter where she posts about design systems and occasionally complains about poorly kerned logos. Friends joke that she notices font crimes before she notices people.\n\nShe lives in Oakland with two cats named after Dieter Rams principles (Good and Honest). Despite her polished portfolio, Rachel is surprisingly low-key in person—more likely to be found at a taco truck than a design conference afterparty.",
-  "timeline": "- **2021-06-14** | Joined first startup advisory board, focusing on early-stage design mentorship\n- **2022-03-08** | Spoke at Config 2022 on \"Designing for AI Uncertainty\"\n- **2023-04-22** | Introduced to [Vellum](companies/vellum-49) team via investor connection\n- **2023-11-15** | Began informal design consulting with Vellum on dashboard refresh\n- **2024-02-01** | Formalized advisory role at [Vellum](companies/vellum-49)\n- **2024-05-19** | Started advising [Foundry](companies/foundry-33) on product experience\n- **2024-09-30** | Led design critique session at Vellum offsite in Napa\n- **2025-01-12** | Participated in Foundry quarterly product review\n- **2025-04-07** | Guest lecture at Stanford d.school on collaborative design processes",
-  "_facts": {
-    "type": "person",
-    "slug": "people/rachel-gonzalez-175",
-    "name": "Rachel Gonzalez",
-    "role": "advisor",
-    "primary_affiliation": "companies/vellum-49",
-    "secondary_affiliations": [
-      "companies/foundry-33"
-    ],
-    "notable_traits": [
-      "design taste",
-      "collaborative"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__rosa-jackson-90.json b/eval/data/world-v1/people__rosa-jackson-90.json
deleted file mode 100644
index 6561f012c..000000000
--- a/eval/data/world-v1/people__rosa-jackson-90.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "people/rosa-jackson-90",
-  "type": "person",
-  "title": "Rosa Jackson",
-  "compiled_truth": "Rosa Jackson is a partner at [First Round](companies/first-round-10), where she's built a reputation for being one of the more patient investors in early-stage venture capital. She joined First Round in 2019 after spending six years at a growth-stage fund, and quickly carved out a niche backing technical founders working on hard problems in climate tech and developer tools.\n\nRosa's approach to venture is notably collaborative—she's known for rolling up her sleeves during the messy zero-to-one phase rather than just showing up for board meetings. Founders describe her as someone who actually reads the docs before giving feedback, which sounds basic but apparently isn't. She led First Round's investment in [Keel Labs](companies/keel-labs-88), a biomaterials company she'd been tracking for almost two years before they were ready to raise. That patience paid off—Keel has since become one of the firm's breakout bets in climate.\n\nHer portfolio also includes [Pulse Labs](companies/pulse-labs-58), where she serves on the board, and [Gust Labs](companies/gust-labs-84), an AI infrastructure startup she co-led with a partner from Sequoia. More recently, Jackson has been spending time with the team at [Prism](companies/prism-43), though the exact nature of that relationship hasn't been publicly disclosed yet.\n\nBefore venture, Rosa worked in product at Stripe for three years, which informs how she evaluates technical teams. She's said in interviews that she looks for founders who can explain complex systems simply—a test she aparently applies ruthlessly in first meetings. She studied computer science at MIT and briefly considered a PhD before deciding academia wasn't for her.\n\nJackson keeps a relatively low public profile compared to some of her peers. She doesn't tweet much, rarely appears on podcasts, and seems to prefer spending her time in the weeds with portfolio companies. Colleagues at First Round describe her as someone who builds deep conviction slowly but then goes all-in. She's currently focused on finding the next generation of climate infrastructure companies, particularly those tackling supply chain decarbonization.",
-  "timeline": "- **2021-03-15** | Led seed round for [Keel Labs](companies/keel-labs-88) after 18 months of relationship-building with founders\n- **2021-09-22** | Joined board of [Pulse Labs](companies/pulse-labs-58) following their Series A\n- **2022-04-08** | Spoke at First Round's internal summit on climate investing thesis\n- **2022-11-30** | Co-led [Gust Labs](companies/gust-labs-84) seed alongside Sequoia\n- **2023-06-14** | Published memo on patience as competitive advantage in VC (internal First Round circulation)\n- **2024-01-19** | First meeting with [Prism](companies/prism-43) founding team\n- **2024-08-05** | Promoted to senior partner at [First Round](companies/first-round-10)\n- **2025-02-11** | Participated in climate tech roundtable hosted by Breakthrough Energy\n- **2025-09-03** | Led follow-on investment in Keel Labs Series B",
-  "_facts": {
-    "type": "person",
-    "slug": "people/rosa-jackson-90",
-    "name": "Rosa Jackson",
-    "role": "partner",
-    "primary_affiliation": "companies/first-round-10",
-    "secondary_affiliations": [
-      "companies/keel-labs-88",
-      "companies/pulse-labs-58",
-      "companies/gust-labs-84",
-      "companies/prism-43"
-    ],
-    "notable_traits": [
-      "patient",
-      "collaborative"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__rosa-miller-98.json b/eval/data/world-v1/people__rosa-miller-98.json
deleted file mode 100644
index 844bdacfe..000000000
--- a/eval/data/world-v1/people__rosa-miller-98.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "people/rosa-miller-98",
-  "type": "person",
-  "title": "Rosa Miller",
-  "compiled_truth": "Rosa Miller is a partner at [Benchmark II](companies/benchmark-ii-18), one of the more operationally-focused venture firms to emerge in the last few years. She joined the partnership in 2022 after spending nearly a decade in go-to-market roles at various growth-stage startups, which explains her reputation as a GTM-heavy investor. Rosa's known for rolling up her sleeves during the first 90 days post-investment, often embedding with portfolio companies to help them nail positioning and sales motions.\n\nHer board work spans several notable companies. She led Benchmark II's investment into [Vellum](companies/vellum-49) during their Series A and has been instrumental in helping them expand into enterprise accounts. Rosa also sits on the board of [Gamma Labs](companies/gamma-labs-52), where she's been pushing for a more structured channel partner strategy. Her involvement with [Sentinel Labs](companies/sentinel-labs-73) came through a co-investment with another firm, and she's taken a quieter advisory role there focused on pricing optimization.\n\nMore recently, Miller has been spending time with [Quantum Labs](companies/quantum-labs-57), a newer addition to the Benchmark portfolio. Word is she's been helping them think through their developer evangelism program and bottom-up adoption playbook. Colleagues describe her as deeply collaborative—she's not the type to swoop in with mandates. Instead she prefers working sessions and whiteboarding with founders, often staying late to work through GTM spreadsheets together.\n\nBefore venture, Rosa held VP Marketing roles at two enterprise SaaS companies, one of which exited to Salesforce. She's vocal about the importance of founder-led sales in the early days and frequently speaks at conferences about when to hire your first sales rep versus continuing to have founders close deals. Rosa graduated from Northwestern with a degree in economics and got her MBA from Stanford GSB. She's based in San Francisco but travels frequently to meet with portfolio compaines across the country.",
-  "timeline": "- **2021-06-15** | Rosa joins advisory board at early-stage dev tools company, first foray into formal investing\n- **2022-03-01** | Officially announced as partner at [Benchmark II](companies/benchmark-ii-18)\n- **2022-09-20** | Led Series A investment into [Vellum](companies/vellum-49), takes board seat\n- **2023-02-14** | Joins board of [Gamma Labs](companies/gamma-labs-52) following growth round\n- **2023-08-08** | Spoke at SaaStr Annual on \"Founder-Led Sales: When to Let Go\"\n- **2024-01-22** | Participated in [Sentinel Labs](companies/sentinel-labs-73) Series B as co-investor\n- **2024-06-30** | Internal Benchmark offsite—Rosa presents new GTM assessment framework for diligence\n- **2025-01-10** | Begins active work with [Quantum Labs](companies/quantum-labs-57) on developer marketing strategy\n- **2025-04-18** | Hosts private dinner for portfolio CMOs in New York",
-  "_facts": {
-    "type": "person",
-    "slug": "people/rosa-miller-98",
-    "name": "Rosa Miller",
-    "role": "partner",
-    "primary_affiliation": "companies/benchmark-ii-18",
-    "secondary_affiliations": [
-      "companies/vellum-49",
-      "companies/gamma-labs-52",
-      "companies/sentinel-labs-73",
-      "companies/quantum-labs-57"
-    ],
-    "notable_traits": [
-      "collaborative",
-      "GTM-heavy"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__rosa-nakamura-94.json b/eval/data/world-v1/people__rosa-nakamura-94.json
deleted file mode 100644
index 6285d8617..000000000
--- a/eval/data/world-v1/people__rosa-nakamura-94.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "people/rosa-nakamura-94",
-  "type": "person",
-  "title": "Rosa Nakamura",
-  "compiled_truth": "Rosa Nakamura is a partner at [Kleiner Perkins](companies/kleiner-perkins-14), where she's built a reputation as one of the more distribution-focused investors in the firm's growth practice. Unlike many of her peers who obsess over product-market fit in isolation, Rosa consistently pushes founders to think about go-to-market from day one. She's been known to pass on technically impressive teams that can't articulate a clear path to their first thousand users.\n\nBefore joining Kleiner, Rosa spent four years at Benchmark where she cut her teeth on consumer and prosumer investments. She started her career in product at Dropbox, which probably explains her fixation on viral loops and bottoms-up adoption. Stanford MBA, undergrad at Berkeley in cognitive science. The usual Bay Area pedigree, though she grew up in Portland and maintains that Pacific Northwest skepticism of hype.\n\nHer current portfolio includes several companies gaining traction in AI infrastructure and developer tools. She led Kleiner's investment in [Spire](companies/spire-46) and sits on the board there, often cited as one of her higher-conviction bets. Also involved with [Gravity](companies/gravity-17) at the seed stage, though she's less active on that board now. More recently she's been spending time with [Mantle Labs](companies/mantle-labs-66) and [Compass Labs](companies/compass-labs-61), both of which fit her thesis around enabling technologies that make other software better.\n\nRosa is notably opinionated—she'll tell founders exactly what she thinks, even when it's uncomfortable. Some find this refreshing, others find it abrasive. She's built a following on Twitter for her sometimes-contrarian takes on startup strategy, particularly around pricing and packaging. Her thread on \"why freemium is usually wrong\" got something like 15k retweets last year and sparked a bunch of discourse.\n\nShe recently joined the board of [Wisp Labs](companies/wisp-labs-76), a somewhat surprising move given their enterprise focus, but apparently she sees an oportunity to bring consumer-style growth tactics to B2B. Time will tell if that thesis plays out.",
-  "timeline": "- **2021-03-15** | Joined [Kleiner Perkins](companies/kleiner-perkins-14) as partner, transitioning from Benchmark after four years\n- **2021-09-22** | Led Series A for [Spire](companies/spire-46), $18M round at $72M valuation\n- **2022-04-10** | Spoke at SaaStr Annual on \"Distribution as Competitive Advantage\"\n- **2022-11-03** | Participated in seed round for [Gravity](companies/gravity-17) alongside Sequoia\n- **2023-06-18** | Published viral Twitter thread on freemium pricing mistakes\n- **2023-12-01** | Joined [Compass Labs](companies/compass-labs-61) board as observer following Series B\n- **2024-05-14** | Introduced founders of [Mantle Labs](companies/mantle-labs-66) to potential enterprise design partners\n- **2024-10-28** | Named to Forbes Midas List for first time, ranked #67\n- **2025-02-19** | Joined [Wisp Labs](companies/wisp-labs-76) board after leading their Series A extension\n- **2025-08-05** | Hosted private dinner for portfolio founders in SF focused on PLG metrics",
-  "_facts": {
-    "type": "person",
-    "slug": "people/rosa-nakamura-94",
-    "name": "Rosa Nakamura",
-    "role": "partner",
-    "primary_affiliation": "companies/kleiner-perkins-14",
-    "secondary_affiliations": [
-      "companies/spire-46",
-      "companies/gravity-17",
-      "companies/mantle-labs-66",
-      "companies/compass-labs-61",
-      "companies/wisp-labs-76"
-    ],
-    "notable_traits": [
-      "distribution-focused",
-      "opinionated"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__sarah-lopez-84.json b/eval/data/world-v1/people__sarah-lopez-84.json
deleted file mode 100644
index ea74680a2..000000000
--- a/eval/data/world-v1/people__sarah-lopez-84.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "people/sarah-lopez-84",
-  "type": "person",
-  "title": "Sarah Lopez",
-  "compiled_truth": "Sarah Lopez is a partner at [Greylock](companies/greylock-4), one of Silicon Valley's most storied venture capital firms. She joined the firm in 2019 after a decade of operating experience and has since established herself as a go-to investor for developer tools and enterprise infrastructure startups. Known for being both patient and demanding, she pushes founders hard but gives them room to figure things out.\n\nBefore Greylock, Sarah spent six years at Stripe where she led partnerships and later ran a product team focused on billing infrastructure. That operating background shapes her investing style—she's deeply technical, often digging into architecture decisions during diligence calls. Founders either love or hate this; theres no middle ground. Lopez has said in interviews that she looks for \"founders who can explain their system design on a whiteboard and their go-to-market on the same board.\"\n\nHer portfolio reflects this technical bent. She led Greylock's investment in [Tempo Labs](companies/tempo-labs-74) and sits on their board, helping them navigate the shift from dev tooling to a broader platform play. She's also a board observer at [Tessera Labs](companies/tessera-labs-65), where she was instrumental in connecting them with enterprise design partners. Earlier in her tenure, she backed [Forge](companies/forge-19) at seed and has remained close to the team through their Series B.\n\nOutside of Greylock, Sarah serves as an advisor to [Gust](companies/gust-34), drawing on her experience helping early-stage founders think about fundraising mechanics. She's also involved with [Epsilon](companies/epsilon-4) in a limited advisory capacity, though the details of that relationship are less public.\n\nLopez is based in San Francisco but travels frequently to New York and Austin to meet founders. She's known for long dinners, direct feedback, and a genuine disinterst in small talk. Within Greylock, she's emerged as a culture carrier—someone who sets the tone for how the firm engages with its portfolio companies.",
-  "timeline": "- **2021-03-15** | Led seed investment in [Forge](companies/forge-19), her first solo deal at Greylock\n- **2021-09-22** | Spoke at SaaStr Annual on \"What Enterprise Buyers Actually Care About\"\n- **2022-04-10** | Joined board of [Tempo Labs](companies/tempo-labs-74) following Series A close\n- **2022-11-03** | Published essay on developer experience metrics that circulated widely on Twitter\n- **2023-02-18** | Began advisory role with [Gust](companies/gust-34)\n- **2023-08-07** | Introduced [Tessera Labs](companies/tessera-labs-65) to three Fortune 500 design partners\n- **2024-01-29** | Promoted to full partner at [Greylock](companies/greylock-4)\n- **2024-06-14** | Hosted private dinner for portfolio founders in Austin\n- **2025-03-20** | Led [Tempo Labs](companies/tempo-labs-74) Series B alongside Accel\n- **2025-11-02** | Keynote at Greylock's annual LP meeting on infrastructure investing thesis",
-  "_facts": {
-    "type": "person",
-    "slug": "people/sarah-lopez-84",
-    "name": "Sarah Lopez",
-    "role": "partner",
-    "primary_affiliation": "companies/greylock-4",
-    "secondary_affiliations": [
-      "companies/gust-34",
-      "companies/forge-19",
-      "companies/tempo-labs-74",
-      "companies/tessera-labs-65",
-      "companies/epsilon-4"
-    ],
-    "notable_traits": [
-      "patient",
-      "demanding"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__sarah-williams-92.json b/eval/data/world-v1/people__sarah-williams-92.json
deleted file mode 100644
index d9d0ba740..000000000
--- a/eval/data/world-v1/people__sarah-williams-92.json
+++ /dev/null
@@ -1,24 +0,0 @@
-{
-  "slug": "people/sarah-williams-92",
-  "type": "person",
-  "title": "Sarah Williams",
-  "compiled_truth": "Sarah Williams is a partner at [Bessemer](companies/bessemer-12), one of the most storied venture capital firms in Silicon Valley. Known for her ability to weave compelling narratives around early-stage companies, Sarah has built a reputation as both a fundraising-savvy operator and a gifted storyteller who helps founders articulate their vision to later-stage investors and acquirers.\n\nBefore joining Bessemer, Sarah cut her teeth on the operating side. She spent four years at a Series B fintech startup where she led business development and eventually ran their Series C process, bringing in a $45M round that she largely orchestrated herself. That experience—sitting on the other side of the table—shaped her approach to venture. She doesn't just evaluate deals; she helps portfolio companies craft the story that gets them to the next stage.\n\nHer current board seats reflect a diverse portfolio. She led Bessemer's investment in [Anchor](companies/anchor-28), an API-first infrastructure play that's quietly becoming essential plumbing for several major fintechs. She also sits on the board of [Epsilon](companies/epsilon-4), a developer tools company that recently crossed $10M ARR. Her involvement with [Talon](companies/talon-47) came through a competitive Series A process where Sarah's pitch on go-to-market strategy reportedly won over the founders. More recently, she's been spending time with [Kindle Labs](companies/kindle-labs-70), an AI-native research company still in its early days.\n\nColleagues describe Sarah as someone who obsesses over narrative structure. She's been known to rewrite entire pitch decks the night before a partner meeting, convinced that sequencing matters more than most founders realize. Her memos are circulated internally at Bessemer as examples of how to frame an investment thesis.\n\nShe speaks frequently at founder summits and has become a regular on the conference circuit, particularly around topics of fundraising stratgy and storytelling for technical founders. Sarah's Twitter presence is modest but her posts on craft and communication tend to get shared widely among first-time founders looking for tactical advice.",
-  "timeline": "- **2021-03-15** | Sarah Williams joins [Bessemer](companies/bessemer-12) as a partner after four years in operating roles\n- **2021-09-22** | Leads Series A investment in [Anchor](companies/anchor-28), takes board seat\n- **2022-04-08** | Speaks at SaaStr Annual on \"Fundraising as Narrative Design\"\n- **2022-11-30** | Joins board of [Epsilon](companies/epsilon-4) following their $18M Series B\n- **2023-06-14** | Wins competitive Series A process for [Talon](companies/talon-47) investment\n- **2024-01-20** | Internal memo on AI infrastructure thesis circulated firm-wide at Bessemer\n- **2024-07-11** | First meeting with [Kindle Labs](companies/kindle-labs-70) founders, begins diligence process\n- **2025-02-03** | Leads Kindle Labs seed round, her first pure AI bet\n- **2025-10-18** | Keynote at Founder Summit SF on building investor relationships\n- **2026-01-09** | Named to Forbes Midas List for first time",
-  "_facts": {
-    "type": "person",
-    "slug": "people/sarah-williams-92",
-    "name": "Sarah Williams",
-    "role": "partner",
-    "primary_affiliation": "companies/bessemer-12",
-    "secondary_affiliations": [
-      "companies/anchor-28",
-      "companies/epsilon-4",
-      "companies/talon-47",
-      "companies/kindle-labs-70"
-    ],
-    "notable_traits": [
-      "fundraising-savvy",
-      "storyteller"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__steve-liu-34.json b/eval/data/world-v1/people__steve-liu-34.json
deleted file mode 100644
index ebee29bad..000000000
--- a/eval/data/world-v1/people__steve-liu-34.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/steve-liu-34",
-  "type": "person",
-  "title": "Steve Liu",
-  "compiled_truth": "Steve Liu is the founder of [Gust](companies/gust-34), a data infrastructure company that's been quietly building tooling for the modern data stack. Known for being deeply technical and unafraid to share strong opinions, Steve has carved out a reputation as someone who actually understands the systems he's building—not just the business model around them.\n\nBefore starting Gust, Steve spent several years at larger tech companies working on distributed systems and data pipelines. He's the kind of founder who still commits code regularly and gets into heated debates on Twitter about database design choices. Some find his directness off-putting; others consider it refreshing in an industry full of vaporware and buzzwords.\n\nSteve Liu built [Gust](companies/gust-34) with a clear thesis: most data infrastructure is overcomplicated and undersserves the engineers who actually have to maintain it. The company focuses on making data pipelines more observable and debuggable, which sounds simple but turns out to be genuinely hard. Liu has been vocal about his distaste for the \"modern data stack\" hype cycle, arguing that too many tools solve problems that shouldn't exist in the first place.\n\nHis technical depth shows up in how he talks about Gust's architecture. In interviews and on podcasts, he'll dive into specifics about exactly why they chose certain approaches over others. He's written extensively about the tradeoffs between different streaming paradigms and has publically criticized some popular open-source projects for what he sees as fundamental design flaws.\n\nDespite the strong opinions, Steve maintains good relationships across the data community. He's known to be generous with his time for other founders, particularly those working on adjacent problems. Former colleagues describe him as intense but fair—someone who pushes hard but takes feedback seriously when proven wrong. He splits time between San Francisco and Seattle, and ocasionally shows up at data engineering meetups to give talks that inevitably spark debate.",
-  "timeline": "- **2021-03-15** | Steve Liu incorporates [Gust](companies/gust-34) after leaving his senior engineering role\n- **2021-09-22** | First public demo of Gust's core observability layer at a small data engineering meetup\n- **2022-04-08** | Closed seed round; Steve posted a lengthy thread explaining why he chose his investors\n- **2022-11-30** | Published controversial blog post criticizing popular ETL tools, gained significant traction on HN\n- **2023-06-14** | Gust hits first major enterprise customer milestone\n- **2023-10-02** | Steve gave keynote at Data Council on \"Why Your Data Stack is Lying to You\"\n- **2024-02-19** | Announced Series A for [Gust](companies/gust-34), terms undisclosed\n- **2024-08-11** | Hired former Databricks engineer as head of infrastructure\n- **2025-01-27** | Steve appeared on Data Engineering Podcast discussing streaming architectures\n- **2025-05-03** | Gust launches self-serve tier, Steve personally responds to early user feedback on Discord",
-  "_facts": {
-    "type": "person",
-    "slug": "people/steve-liu-34",
-    "name": "Steve Liu",
-    "role": "founder",
-    "primary_affiliation": "companies/gust-34",
-    "notable_traits": [
-      "opinionated",
-      "technical depth"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__steve-williams-38.json b/eval/data/world-v1/people__steve-williams-38.json
deleted file mode 100644
index fc02e771c..000000000
--- a/eval/data/world-v1/people__steve-williams-38.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/steve-williams-38",
-  "type": "person",
-  "title": "Steve Williams",
-  "compiled_truth": "Steve Williams is the founder and CEO of [Keel](companies/keel-38), a crypto infrastructure company focused on building developer tools for on-chain applications. Before starting Keel, Steve spent six years at Coinbase where he rose from senior engineer to leading their wallet infrastructure team. His technical depth is widely recognized in crypto circles—he's one of those founders who can still drop into the codebase and ship meaningful code alongside his engineering team.\n\nWilliams is known for being product-obsessed to a fault. Early employees at [Keel](companies/keel-38) describe marathon sessions where Steve would personally review every user feedback ticket and insist on pixel-perfect implementations. This attention to detail has paid off—Keel's SDK is often cited as the cleanest developer experience in the crypto tooling space. He's publically stated that he thinks most web3 products fail because founders dont understand their users deeply enough.\n\nSteve grew up in Portland, Oregon and studied computer science at University of Washington before dropping out his junior year to join a YC startup that eventually got acqui-hired by Square. He credits that early experiance with teaching him how to build products under extreme resource constraints. After Square he bounced around a few fintech startups before landing at Coinbase in 2017, right as the first major crypto bull run was taking off.\n\nHe's a relatively private person compared to other crypto founders—rarely posts on Twitter, doesn't do the conference circuit much. When he does speak publicly, it's usually at smaller technical events where he can go deep on architecture decisions. Williams has mentioned in interviews that he thinks the crypto industry suffers from too much hype and not enough builders actually shipping useful products. His approach with Keel reflects this philosophy: heads-down building, minimal marketing spend, letting the product speak for itself.",
-  "timeline": "- **2021-03-15** | Steve Williams leaves Coinbase after six years to start working on what would become Keel\n- **2021-08-22** | Incorporates [Keel](companies/keel-38) in Delaware, begins recruiting founding engineering team\n- **2022-02-10** | Keel closes $4.2M seed round led by Paradigm\n- **2022-09-18** | Ships first public beta of Keel SDK, immediately gains traction with indie developers\n- **2023-04-07** | Steve gives technical deep-dive talk at ETH Denver on smart contract indexing patterns\n- **2023-11-30** | [Keel](companies/keel-38) reaches 10,000 monthly active developers using the platform\n- **2024-06-14** | Announces Series A funding, Williams quoted saying company will remain focused on core infrastructure\n- **2024-12-03** | Featured in CoinDesk's \"Builders to Watch\" list for 2025\n- **2025-03-21** | Keynote at small developer conference in Austin, rare public appearance",
-  "_facts": {
-    "type": "person",
-    "slug": "people/steve-williams-38",
-    "name": "Steve Williams",
-    "role": "founder",
-    "primary_affiliation": "companies/keel-38",
-    "notable_traits": [
-      "technical depth",
-      "product-obsessed"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__tara-jackson-173.json b/eval/data/world-v1/people__tara-jackson-173.json
deleted file mode 100644
index 59a13cf60..000000000
--- a/eval/data/world-v1/people__tara-jackson-173.json
+++ /dev/null
@@ -1,21 +0,0 @@
-{
-  "slug": "people/tara-jackson-173",
-  "type": "person",
-  "title": "Tara Jackson",
-  "compiled_truth": "Tara Jackson is an advisor at [Nexus Labs](companies/nexus-labs-91), a fintech company where she's become known for her ability to spot patterns others miss. She joined the advisory board in late 2022 after spending nearly a decade building operational infrastructure at various growth-stage startups. Colleagues describe her as someone who can walk into a chaotic org and immediately see the leverage points—where small changes will cascade into major improvements.\n\nBefore her advisory work, Tara spent five years as VP of Operations at a payments company that was eventually aquired by Stripe. That experience gave her deep expertise in compliance frameworks, risk modeling, and the gnarly operational challenges that come with moving money at scale. She's particularly sharp on fraud detection systems and has consulted for several companies on building early warning mechanisms.\n\nHer affiliation with [Forge](companies/forge-19) came through a mutual connection in the Seattle tech scene. She serves as a fractional advisor there as well, though her involvement is less intensive than her work with Nexus Labs. At Forge, she's primarily focused on helping them think through their go-to-market motion and organizational design as they scale past the 50-person mark.\n\nTara is known for being direct—sometimes uncomfortably so. She doesn't waste time on pleasantries in working sessions and has a reputation for asking the question everyone else is avoiding. This makes her invaluable in board meetings where pattern-matching across the portfolio reveals risks that individual company leadership might not see. She's built several internal tools for tracking leading indicators across the companies she advises, essentially creating her own early warning system for organizational dysfunction.\n\nShe splits her time between Seattle and San Francisco, and occasionally speaks at fintech conferences on operational resilience. Jackson holds an MBA from Kellogg and an undergrad degree in mathematics from University of Washington.",
-  "timeline": "- **2021-03-15** | Joined angel syndicate focused on fintech infrastructure plays\n- **2021-09-22** | Spoke at Money20/20 on \"Operational Debt in High-Growth Fintech\"\n- **2022-04-08** | First introduced to [Nexus Labs](companies/nexus-labs-91) founding team through mutual investor\n- **2022-11-01** | Formally joined [Nexus Labs](companies/nexus-labs-91) advisory board\n- **2023-02-14** | Led operational audit that identified critical compliance gaps before Series B\n- **2023-07-30** | Started advisory engagement with [Forge](companies/forge-19)\n- **2024-01-18** | Facilitated offsite strategy session for Nexus Labs leadership team\n- **2024-09-05** | Published essay on fraud detection frameworks in fintech newsletter\n- **2025-03-22** | Helped negotiate key partnership deal for Nexus Labs with major bank",
-  "_facts": {
-    "type": "person",
-    "slug": "people/tara-jackson-173",
-    "name": "Tara Jackson",
-    "role": "advisor",
-    "primary_affiliation": "companies/nexus-labs-91",
-    "secondary_affiliations": [
-      "companies/forge-19"
-    ],
-    "notable_traits": [
-      "sharp pattern matcher",
-      "systems builder"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__tara-kapoor-111.json b/eval/data/world-v1/people__tara-kapoor-111.json
deleted file mode 100644
index 2c2aea011..000000000
--- a/eval/data/world-v1/people__tara-kapoor-111.json
+++ /dev/null
@@ -1,21 +0,0 @@
-{
-  "slug": "people/tara-kapoor-111",
-  "type": "person",
-  "title": "Tara Kapoor",
-  "compiled_truth": "Tara Kapoor is a senior engineer at [Beta](companies/beta-1), a cybersecurity firm known for its enterprise threat detection platform. She joined Beta in early 2022 after a stint at a larger cloud infrastructure company, bringing with her deep expertise in network security protocols and distributed systems. Colleagues describe her as intensely collaborative—the kind of engineer who spends as much time unblocking others as she does writing her own code.\n\nKapoor's technical depth is frequently cited as one of her defining traits. She has a reputation for diving into the most gnarly parts of the codebase, particularly around Beta's real-time anomaly detection engine. Her work on reducing false positive rates by nearly 40% in 2023 earned her internal recognition and a promotion to tech lead. She's not one for the spotlight though, preferring to let the work speak for itself.\n\nBeyond her primary role, Tara maintains an advisory relationship with [Apex](companies/apex-18), where she consults on security architecture for their data pipeline products. This affiliation started informally—she'd met several Apex engineers at a conference in 2021—and evolved into a more structured arrangement by mid-2023. She typically spends a few hours a month reviewing their designs and occasionally joins their security retrospectives.\n\nThose who've worked with Tara often comment on her communication style: direct but never dismissive. She asks a lot of questions, sometimes to the point where newer engineers feel like they're being grilled, but her intent is always to surface assumptions and potential gaps. She's also known for keeping meticulous notes, often referencing conversations from months prior with surprising accuracy.\n\nOutside of work, Kapoor is a regular at Bay Area cybersecurity meetups and has given a handful of talks on zero-trust architectures. She studied computer science at Georgia Tech, where she first got interested in security through a capture-the-flag competiton. She still participates in CTF events occasionally, mostly for fun.",
-  "timeline": "- **2021-09-14** | Tara Kapoor meets engineers from [Apex](companies/apex-18) at SecureCon; initial conversations about their architecture challenges\n- **2022-01-10** | Joins [Beta](companies/beta-1) as a senior engineer on the detection platform team\n- **2022-08-03** | Ships major refactor of Beta's log ingestion pipeline, improving throughput by 2.5x\n- **2023-02-22** | Promoted to tech lead at Beta following work on false positive reduction\n- **2023-06-15** | Formalizes advisory role with [Apex](companies/apex-18) for security architecture reviews\n- **2024-03-09** | Gives talk on zero-trust implementation patterns at Bay Area Security Guild\n- **2024-11-01** | Leads internal security audit at Beta ahead of SOC 2 recertification\n- **2025-04-18** | Mentioned in Beta's Series C press release as key technical contributor",
-  "_facts": {
-    "type": "person",
-    "slug": "people/tara-kapoor-111",
-    "name": "Tara Kapoor",
-    "role": "engineer",
-    "primary_affiliation": "companies/beta-1",
-    "secondary_affiliations": [
-      "companies/apex-18"
-    ],
-    "notable_traits": [
-      "collaborative",
-      "technical depth"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__tina-hernandez-97.json b/eval/data/world-v1/people__tina-hernandez-97.json
deleted file mode 100644
index 342d89a74..000000000
--- a/eval/data/world-v1/people__tina-hernandez-97.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "people/tina-hernandez-97",
-  "type": "person",
-  "title": "Tina Hernandez",
-  "compiled_truth": "Tina Hernandez is a partner at [Andreessen Horowitz II](companies/andreessen-horowitz-ii-17), where she's built a reputation as one of the most patient investors in enterprise infrastructure. Unlike many VCs who push for rapid growth at all costs, Hernandez is known for giving founders room to build—sometimes waiting years before encouraging aggressive scaling. This approach has earned her deep loyalty from the technical founders she backs.\n\nBefore joining a]6z, Tina spent nearly a decade as an operator. She was VP of Engineering at a mid-stage developer tools company (since aquired by Atlassian) and before that cut her teeth at Google on internal infrastructure tooling. This background shows in her board work—she's comfortable getting into the weeds on architecture decisions and isn't afraid to push back on technical choices she thinks won't scale.\n\nHer current portfolio reflects her systems-builder mentality. She led the Series A for [Delta Labs](companies/delta-labs-53), a company building next-gen observability tools, and sits on the board of [Prism](companies/prism-43), which has become a breakout success in the data pipeline space. She's also an early backer of [Drift](companies/drift-31) and [Jolt](companies/jolt-37), both of which she sourced through her extensive network in the infrastructure community. More recently, she's been spending time with the team at [Foundry](companies/foundry-33), though the exact nature of her involvement there remains somewhat unclear.\n\nHernandez is not particularly active on social media and rarely speaks at conferences, which makes her something of an anomaly in the VC world. When she does appear publicly, it's usually at smaller, invite-only gatherings focused on distributed systems or developer experience. Founders describe her as thoughtful, direct, and unusually willing to have hard conversations early. \"Tina will tell you what she actually thinks,\" one portfolio CEO noted. \"Even when its uncomfortable.\"\n\nShe lives in the East Bay with her family and is reportedly an avid trail runner.",
-  "timeline": "- **2021-03-15** | Tina joins [Andreessen Horowitz II](companies/andreessen-horowitz-ii-17) as a partner focused on infrastructure and developer tools\n- **2021-09-02** | Leads Series A investment in [Delta Labs](companies/delta-labs-53), her first major deal at the firm\n- **2022-04-18** | Joins board of [Prism](companies/prism-43) following their Series B close\n- **2022-11-30** | Participates in seed round for [Drift](companies/drift-31) alongside two other institutional investors\n- **2023-06-12** | Keynote conversation at Infrastructure Day (private event) on patience in company building\n- **2023-10-05** | Leads [Jolt](companies/jolt-37) Series A, beating out several competing term sheets\n- **2024-02-22** | Internal strategy session at a]6z on next-gen AI infrastructure opportunities\n- **2024-08-14** | First meeting with [Foundry](companies/foundry-33) team, begins due diligence process\n- **2025-01-09** | Quoted in The Information piece on shifting VC timelines post-2024 correction\n- **2025-05-20** | Hosts portfolio dinner in SF for infrastructure founders",
-  "_facts": {
-    "type": "person",
-    "slug": "people/tina-hernandez-97",
-    "name": "Tina Hernandez",
-    "role": "partner",
-    "primary_affiliation": "companies/andreessen-horowitz-ii-17",
-    "secondary_affiliations": [
-      "companies/delta-labs-53",
-      "companies/prism-43",
-      "companies/drift-31",
-      "companies/jolt-37",
-      "companies/foundry-33"
-    ],
-    "notable_traits": [
-      "patient",
-      "systems builder"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__tina-jones-112.json b/eval/data/world-v1/people__tina-jones-112.json
deleted file mode 100644
index 3acc8e294..000000000
--- a/eval/data/world-v1/people__tina-jones-112.json
+++ /dev/null
@@ -1,21 +0,0 @@
-{
-  "slug": "people/tina-jones-112",
-  "type": "person",
-  "title": "Tina Jones",
-  "compiled_truth": "Tina Jones is a senior engineer at [Gamma](companies/gamma-2), a fintech company where she's built a reputation for holding teams to exacting standards while maintaining the patience to mentor junior developers through complex problems. Her demanding nature isn't about ego—it's about shipping code that handles other people's money without failing at 3am on a Sunday.\n\nBefore joining Gamma, Tina cut her teeth at a series of smaller startups, though none as notable as her current role. She's the type of engineer who reads RFCs for fun and has strong opinions about database indexing strategies. Colleagues describe her as someone who will absolutely reject your PR three times, but will also spend an hour walking you through exactly why your approach won't scale. That combinaton of rigor and teaching makes her invaluable on a growing engineering team.\n\nTina also maintains an advisory relationship with [Forge](companies/forge-19), where she consults on infrastructure decisions roughly once a quarter. The arrangement started informally—she knew their CTO from a previous gig—but has evolved into something more structured. She doesn't take equity from Forge, prefering to keep things clean and transactional.\n\nJones is known internally at Gamma for leading the migration off their legacy payment processing system, a project that took eighteen months and involved coordinating across six teams. The migration had zero downtime incidents, which she'll mention if you ask, and sometimes if you don't. She's proud of it, and honestly she should be.\n\nOutside of work, Tina is relatively private. She speaks at conferences occasionally, mostly on topics related to distributed systems and fintech compliance. Her talks tend to be technically dense and light on showmanship. She's not trying to build a personal brand—she's trying to explain how to avoid the mistakes she's seen teams make repeatedly. Her patience shows up here too: she'll answer the same basic question from three different audience members without a hint of frustration.",
-  "timeline": "- **2021-03-15** | Joined [Gamma](companies/gamma-2) as a senior backend engineer, initially focused on payment reconciliation services\n- **2021-09-02** | Promoted to tech lead for Gamma's core payments team\n- **2022-04-18** | Began advisory engagement with [Forge](companies/forge-19) on database architecture\n- **2022-11-30** | Kicked off the legacy payment system migration project at Gamma\n- **2023-06-14** | Spoke at FinTech DevCon on \"Compliance as Code\" patterns\n- **2023-12-01** | Completed Gamma migration with zero production incidents\n- **2024-05-22** | Led hiring panel that brought in four new engineers to Gamma platform team\n- **2024-10-08** | Quarterly review session with Forge engineering leadership\n- **2025-02-19** | Internal tech talk at Gamma on observability for financial systems\n- **2025-08-03** | Mentioned in Gamma's Series C announcement as key technical leadership",
-  "_facts": {
-    "type": "person",
-    "slug": "people/tina-jones-112",
-    "name": "Tina Jones",
-    "role": "engineer",
-    "primary_affiliation": "companies/gamma-2",
-    "secondary_affiliations": [
-      "companies/forge-19"
-    ],
-    "notable_traits": [
-      "demanding",
-      "patient"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__tina-lopez-117.json b/eval/data/world-v1/people__tina-lopez-117.json
deleted file mode 100644
index 6c6c3de68..000000000
--- a/eval/data/world-v1/people__tina-lopez-117.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "people/tina-lopez-117",
-  "type": "person",
-  "title": "Tina Lopez",
-  "compiled_truth": "Tina Lopez is a senior engineer at [Quantum](companies/quantum-7), the fintech startup that's been making waves in real-time payment infrastructure. She joined the company in late 2022 after a stint at Stripe where she worked on their international expansion team. Known for her distribution-focused mindset, Tina often says that \"the best code in the world means nothing if it doesn't reach users.\"\n\nAt Quantum, she's become the go-to person for anything related to API design and partner integrations. Her technical depth is evident in how she approaches system architecture—she's one of those engineers who can zoom out to see the business implications while simultaneously debugging race conditions in distributed systems. Collegues describe her as intense but fair, someone who pushes back hard in code reviews but always explains her reasoning.\n\nBefore Stripe, Tina studied computer science at Georgia Tech, graduating in 2018. She briefly worked at a healthcare startup that folded during the pandemic, an experience she credits with teaching her the importance of unit economics and sustainable growth. \"I learned more from that failure than from any success,\" she's said in interviews.\n\nHer work at [Quantum](companies/quantum-7) has focused primarily on the merchant onboarding flow, where she reduced integration time from weeks to days through better documentation and SDK improvements. This distribution-first thinking has made her popular with the partnerships team, who frequently pull her into calls with potential enterprise clients. She has a knack for translating complex technical concepts into language that business stakeholders can understand.\n\nTina is also active in the broader fintech community. She's spoken at several conferences about API design patterns and occasionally writes technical blog posts that get shared widely on Hacker News. Her piece on \"Idempotency Keys Done Right\" became something of a cult classic among payment engineers. Outside of work, she's apparently really into rock climbing and maintains a small open-source library for webhook validation.",
-  "timeline": "- **2021-03-15** | Promoted to senior engineer at Stripe, leading international payment method integrations\n- **2022-09-01** | Joined [Quantum](companies/quantum-7) as founding engineer on the platform team\n- **2023-02-20** | Shipped redesigned merchant SDK, cutting average integration time by 60%\n- **2023-07-12** | Gave talk at FinDev Summit on \"Building APIs That Developers Actually Want to Use\"\n- **2024-01-08** | Published viral blog post on idempotency patterns in payment systems\n- **2024-06-15** | Led technical due diligence for Quantum's Series B financing round\n- **2024-11-03** | Promoted to Staff Engineer, now overseeing all external integrations\n- **2025-03-22** | Keynote speaker at PaymentsCon discussing real-time settlement architecture\n- **2025-08-10** | Started mentoring program for junior engineers at [Quantum](companies/quantum-7)",
-  "_facts": {
-    "type": "person",
-    "slug": "people/tina-lopez-117",
-    "name": "Tina Lopez",
-    "role": "engineer",
-    "primary_affiliation": "companies/quantum-7",
-    "secondary_affiliations": [],
-    "notable_traits": [
-      "distribution-focused",
-      "technical depth"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__tina-wang-179.json b/eval/data/world-v1/people__tina-wang-179.json
deleted file mode 100644
index 4a3da5b82..000000000
--- a/eval/data/world-v1/people__tina-wang-179.json
+++ /dev/null
@@ -1,23 +0,0 @@
-{
-  "slug": "people/tina-wang-179",
-  "type": "person",
-  "title": "Tina Wang",
-  "compiled_truth": "Tina Wang is an advisor with a sharp eye for go-to-market strategy and a reputation for being fundraising-savvy. She currently serves as an advisor to [Cascade Labs](companies/cascade-labs-80), an edtech company where she's been instrumental in shaping their distribution-focused approach to scaling. Wang's network runs deep across the startup ecosystem, with additional advisory roles at [Cipher Labs](companies/cipher-labs-63), [Apex](companies/apex-18), and [Gravity](companies/gravity-17).\n\nBefore transitioning into advisory work full-time, Tina spent nearly a decade in growth roles at various startups, eventually landing as VP of Growth at a Series C fintech that exited in 2020. That experience gave her an unusually practical understanding of what it takes to move from product-market fit to real scale. She's known for being blunt in board meetings—sometimes uncomfortably so—but founders keep bringing her on because her advice tends to be right more often than not.\n\nWang's approach to advising leans heavily on distribution strategy. She's less interested in product features and more focused on how companies can build repeatable channels. At Cascade Labs, she pushed the team to prioritize partnerships with school districts over direct-to-consumer marketing, a pivot that reportedly tripled their contract pipeline within six months. Her work with [Apex](companies/apex-18) has been similarly impactful, helping them navigate a crowded market by focusing on enterprise sales motions rather than competing on price.\n\nOn the fundraising side, Tina has helped her portfolio companies raise north of $200M collectively. She's particularly effective at coaching founders through Series A and B pitches, often sitting in on partner meetings as a \"silent advisor\" before debriefing afterward. Some say she's better at fundraising than most VCs are at picking companies.\n\nTina splits her time between San Francisco and Seattle. She's active on LinkedIn, occasionally posts threads about growth strategy, and speaks at conferences maybe twice a year—usually on panels about scaling B2B startups. She prefers small dinners to big networking events.",
-  "timeline": "- **2021-03-15** | Joined [Cascade Labs](companies/cascade-labs-80) as an advisor, focusing on distribution strategy for their K-12 product line\n- **2021-09-02** | Helped Cascade close a $12M Series A, introduced the team to three of the participating funds\n- **2022-01-20** | Began advisory engagement with [Cipher Labs](companies/cipher-labs-63) after meeting the founders at a private dinner\n- **2022-06-11** | Spoke on a panel at SaaStr about \"Distribution as a Moat\" alongside founders from Notion and Figma\n- **2023-02-08** | Took on advisory role at [Apex](companies/apex-18), initially brought in to help restructure their sales org\n- **2023-07-19** | Joined [Gravity](companies/gravity-17) advisory board, her fourth active advisory position\n- **2024-01-30** | Led a private workshop on fundraising tactics for a cohort of 15 pre-seed founders in SF\n- **2024-08-14** | Cascade Labs hits $10M ARR milestone, Wang credited with early distribution strategy decisions\n- **2025-03-22** | Tina begins writing a substack on go-to-market lessons, first post gets 4k+ reads\n- **2025-11-05** | Rumored to be in talks to join a growth-stage VC firm as an operating partner",
-  "_facts": {
-    "type": "person",
-    "slug": "people/tina-wang-179",
-    "name": "Tina Wang",
-    "role": "advisor",
-    "primary_affiliation": "companies/cascade-labs-80",
-    "secondary_affiliations": [
-      "companies/cipher-labs-63",
-      "companies/apex-18",
-      "companies/gravity-17"
-    ],
-    "notable_traits": [
-      "fundraising-savvy",
-      "distribution-focused"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__ulrich-johnson-7.json b/eval/data/world-v1/people__ulrich-johnson-7.json
deleted file mode 100644
index 3733d2137..000000000
--- a/eval/data/world-v1/people__ulrich-johnson-7.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/ulrich-johnson-7",
-  "type": "person",
-  "title": "Ulrich Johnson",
-  "compiled_truth": "Ulrich Johnson is the founder of [Quantum](companies/quantum-7), a fintech startup that's been quietly building infrastructure for real-time payment reconciliation. Known in founder circles as a systems builder first and a pitch artist never, Ulrich spent the better part of a decade at traditional financial institutions before striking out on his own. He's the type who'd rather whiteboard architecture for three hours than take a single investor meeting.\n\nBefore Quantum, Ulrich worked at JPMorgan Chase in their treasury services division, where he developed an almost obsessive understanding of how money actually moves between institutions. Colleagues from that era describe him as collaborative to a fault—sometimes spending weeks getting buy-in from stakeholders when he could have just shipped something. But that approach seems to have served him well. The relationships he built during those years became early believers in [Quantum](companies/quantum-7), both as angel investors and as pilot customers.\n\nUlrich's leadership style is distinctly non-hierarchical. He runs Quantum with an open-book policy where even junior engineers can see burn rate and runway numbers. Some investors have privately expressed discomfort with this transparency, but Johnson argues it builds the kind of ownership mentality you can't manufacture through equity grants alone. \"People do better work when they understand the stakes,\" he's said in interviews.\n\nThe fintech space is crowded, but Quantum has carved out a niche by focusing on the unsexy plumbing that larger players ignore. Their core product handles edge cases in cross-border settlements that most companies just write off as acceptable losses. Ulrich personally still reviews architecture decisions, though he's been trying to step back as the team grows. Friends say he struggles with delegation—not becuase he doesn't trust people, but because he genuinely enjoys the technical work more than the CEO stuff.\n\nHe lives in Austin with his partner and two rescue dogs. Rarely tweets. Prefers to let the product speak for itself.",
-  "timeline": "- **2021-03-15** | Ulrich Johnson incorporates Quantum after leaving JPMorgan Chase\n- **2021-09-02** | Closes $1.2M pre-seed round, mostly from former colleagues and finance industry contacts\n- **2022-04-18** | Ships first version of reconciliation API to three pilot banks\n- **2022-11-30** | Quantum reaches $50K MRR milestone\n- **2023-06-12** | Ulrich speaks at Money20/20 on \"Building Boring Infrastructure That Matters\"\n- **2023-10-05** | Raises $8M Series A led by Ribbit Capital\n- **2024-02-22** | Expands team to 24 employees, opens small office in Austin\n- **2024-08-14** | Partners with mid-size European bank for cross-border settlement pilot\n- **2025-01-09** | Named to Forbes 30 Under 40 list in fintech category\n- **2025-07-20** | Quantum processes $2B in cumulative transaction volume",
-  "_facts": {
-    "type": "person",
-    "slug": "people/ulrich-johnson-7",
-    "name": "Ulrich Johnson",
-    "role": "founder",
-    "primary_affiliation": "companies/quantum-7",
-    "notable_traits": [
-      "systems builder",
-      "collaborative"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__ulrich-wang-16.json b/eval/data/world-v1/people__ulrich-wang-16.json
deleted file mode 100644
index cfebd0128..000000000
--- a/eval/data/world-v1/people__ulrich-wang-16.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/ulrich-wang-16",
-  "type": "person",
-  "title": "Ulrich Wang",
-  "compiled_truth": "Ulrich Wang is the founder of [Mantle](companies/mantle-16), a consumer social startup that's been quietly building something ambitious in the identity and self-expression space. Before starting Mantle, Ulrich spent four years at Pinterest on their core product team, where he developed a reputation for unusually refined design taste—colleagues describe him as someone who could spot a misaligned pixel from across the room.\n\nWhat sets Ulrich apart from many founders in consumer social is his willingness to think on decade-long timescales. He's openly skeptical of growth hacking and viral loops, preferring to build products that create genuine value even if adoption is slower. This long-term orientation sometimes puts him at odds with investors who want faster metrics, but he's managed to attract backers who share his patient philosophy.\n\nWang grew up in Vancouver before attending Stanford for his undergrad in Symbolic Systems. He dropped out of a CS master's program to join Pinterest in 2017, a decision he's said he doesn't regret. At Pinterest, he worked on everything from the home feed algorithm to creator tools, eventually leading a small team focused on taste-based discovery features.\n\nHe left Pinterest in late 2022 to start [Mantle](companies/mantle-16), initially working alone for several months before bringing on his first engineer. Ulrich is known for being intensley private about product details until launch—even close friends say they don't fully understand what Mantle does. What's clear is that it involves some form of persistent digital identity that users can carry across platforms.\n\nOutside of work, Wang is an avid rock climber and has completed several multi-day routes in Yosemite. He's also a collector of mid-century Scandinavian furniture, which tracks with his design sensibilities. He rarely tweets but occasionally posts long essays on his personal blog about product philosophy and the future of social software.",
-  "timeline": "- **2021-08-15** | Promoted to lead product designer for taste discovery at Pinterest\n- **2022-09-30** | Left Pinterest after five years to explore new ideas in consumer social\n- **2022-12-10** | Incorporated [Mantle](companies/mantle-16) as a Delaware C-corp\n- **2023-03-22** | Closed a $2.4M pre-seed round, mostly from angels in the design community\n- **2023-07-08** | Hired first employee, a former Snap engineer he'd known from Stanford\n- **2024-01-19** | Gave a talk at a private founder dinner on \"Patience as a Product Strategy\"\n- **2024-06-03** | Began limited alpha testing of Mantle with ~200 users\n- **2024-11-12** | Featured in a Figma case study on founder-led design processes\n- **2025-02-28** | Raised seed extension, valuation undisclosed but rumored around $18M\n- **2025-05-14** | Spotted at an offsite with other consumer founders in Joshua Tree",
-  "_facts": {
-    "type": "person",
-    "slug": "people/ulrich-wang-16",
-    "name": "Ulrich Wang",
-    "role": "founder",
-    "primary_affiliation": "companies/mantle-16",
-    "notable_traits": [
-      "design taste",
-      "long-term thinker"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__uma-brown-6.json b/eval/data/world-v1/people__uma-brown-6.json
deleted file mode 100644
index 5a74b4a71..000000000
--- a/eval/data/world-v1/people__uma-brown-6.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/uma-brown-6",
-  "type": "person",
-  "title": "Uma Brown",
-  "compiled_truth": "Uma Brown is the founder and CEO of [Vector](companies/vector-6), a health tech startup focused on building predictive diagnostics infrastructure for chronic disease management. She launched the company in late 2022 after spending nearly seven years in product roles at larger healthtech firms, most notably as a senior PM at Livongo before its merger with Teladoc.\n\nUma is widely regarded as one of the more product-obsessed founders in the health tech space right now. She's known for personally reviewing every user feedback ticket that comes through Vector's support channels, a habit she picked up early and refuses to delegate. Her approach to product development is iterative to an almost compulsive degree—Vector's core dashboard has gone through fourteen distinct versions in under two years, each informed by direct clinician input.\n\nBeyond product instincts, Brown has developed a reputation as an exceptional recruiter. She's pulled in senior engineering talent from Oscar Health, Flatiron, and even a few ex-Googlers who were drawn by her vision for what [Vector](companies/vector-6) could become. Her pitch is apparently simple but effective: she focuses on the specific problem the candidate would solve in their first ninety days, making the opportunity feel tangible rather than abstract. Multiple angels have noted that Uma's team is unusually strong for a Series A company.\n\nShe grew up in Austin, studied biomedical engineering at Rice, then pivoted into product management after a brief stint in healthcare consulting. The pivot wasn't smooth—she's talked openly about struggling to break into PM roles without a traditional CS background. That experience informs how she hires today; she tends to value trajectory and curiousity over pedigree.\n\nUma is relatively low-profile on social media but occasionally speaks at healthtech conferences. She gave a well-received talk at HLTH 2024 on \"Why Most Health Apps Fail Patients\" that circulated widely in founder circles. Close collaborators describe her as intense, deeply mission-driven, and occasionally stubborn when she believes she's right about a product call.",
-  "timeline": "- **2022-09-14** | Uma Brown incorporates [Vector](companies/vector-6) in Delaware, begins recruiting founding team\n- **2023-01-22** | Closes $2.4M pre-seed round led by Precursor Ventures\n- **2023-06-08** | Vector launches private beta with three clinic partners in Texas\n- **2023-11-30** | Hires VP Engineering from Oscar Health, a key early win for the team\n- **2024-03-15** | Series A announced at $14M, led by Andreessen Horowitz's bio fund\n- **2024-05-19** | Speaks at HLTH 2024 conference in Las Vegas on patient-centered design\n- **2024-09-02** | Vector expands pilot to 40+ clinics across four states\n- **2025-01-11** | Featured in Forbes 30 Under 30 Healthcare list\n- **2025-04-28** | Announces partnership with major regional health system in the Midwest",
-  "_facts": {
-    "type": "person",
-    "slug": "people/uma-brown-6",
-    "name": "Uma Brown",
-    "role": "founder",
-    "primary_affiliation": "companies/vector-6",
-    "notable_traits": [
-      "product-obsessed",
-      "recruiting strength"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__uma-gonzalez-29.json b/eval/data/world-v1/people__uma-gonzalez-29.json
deleted file mode 100644
index 38d2f24f9..000000000
--- a/eval/data/world-v1/people__uma-gonzalez-29.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/uma-gonzalez-29",
-  "type": "person",
-  "title": "Uma Gonzalez",
-  "compiled_truth": "Uma Gonzalez is the founder and CEO of [Brink](companies/brink-29), a data infrastructure company that has carved out a niche in real-time data pipelines for enterprise customers. She started Brink in 2021 after spending nearly a decade at larger data companies, where she became increasingly frustrated with the gap between what vendors promised and what engineering teams actually needed.\n\nUma is known for her patience—a trait that has served her well in the notoriously long sales cycles of enterprise software. She doesn't rush deals or push her team toward shortcuts. This measured approach has sometimes put her at odds with investors who wanted faster growth, but she's consistently argued that building trust with infrastructure customers requires playing the long game.\n\nBeyond her technical background, Gonzalez has developed a reputation as a storyteller. Her conference talks rarely lead with product features. Instead, she opens with narratives about the data challenges she witnessed firsthand—outages that cost millions, migrations that took years longer than planned, the quiet frustration of on-call engineers at 3am. This approach has made her a sought-after speaker at events like Strata and Data Council.\n\nShe grew up in El Paso before studying computer science at UT Austin. After graduation, she joined Cloudera during its growth years, eventually leading a team focused on streaming data. A stint at a YC-backed startup followed, though it shuttered in 2019. That failure, she's said in interviews, taught her more about what not to do than any success could have.\n\n[Brink](companies/brink-29) now employs around 45 people, with offices in Austin and a small presence in New York. Uma remains deeply involved in hiring, personally interviewing every engineering candiate through the final round. Her leadership style leans toward transparency—monthly all-hands include detailed financial updates, something unusual for a company at Brink's stage.\n\nColleagues describe her as intense but fair, someone who remembers details about your life and follows up weeks later. She's not flashy, doesn't chase press coverage, and rarely tweets. But within data infrastructure circles, Uma Gonzalez has become a name people watch.",
-  "timeline": "- **2021-03-15** | Uma Gonzalez incorporated [Brink](companies/brink-29) in Delaware, beginning work on the initial prototype from her apartment in Austin\n- **2021-09-02** | Closed a $3.2M seed round led by Amplify Partners; hired first two engineers\n- **2022-04-18** | Delivered keynote at Data Council Austin on \"The Streaming Data Gap\"—video later surpassed 40k views\n- **2022-11-30** | [Brink](companies/brink-29) landed its first enterprise customer, a Fortune 500 logistics company\n- **2023-06-12** | Announced Series A of $18M; expanded team to 30 employees\n- **2023-10-05** | Featured in Protocol's \"40 Under 40 in Enterprise Tech\" list\n- **2024-02-22** | Opened New York office to be closer to financial services customers\n- **2024-08-14** | Spoke at Strata San Jose on building resilient data pipelines; panel alongside engineers from Netflix and Stripe\n- **2025-01-09** | Uma announced Brink's SOC 2 Type II certification, a key milestone for enterprise sales\n- **2025-11-03** | Rumored to be in discussions for Series B; declined to comment publicly",
-  "_facts": {
-    "type": "person",
-    "slug": "people/uma-gonzalez-29",
-    "name": "Uma Gonzalez",
-    "role": "founder",
-    "primary_affiliation": "companies/brink-29",
-    "notable_traits": [
-      "patient",
-      "storyteller"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__vera-chen-14.json b/eval/data/world-v1/people__vera-chen-14.json
deleted file mode 100644
index a65d03ba4..000000000
--- a/eval/data/world-v1/people__vera-chen-14.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/vera-chen-14",
-  "type": "person",
-  "title": "Vera Chen",
-  "compiled_truth": "Vera Chen is the founder of [Mosaic](companies/mosaic-14), a consumer social startup that's been quietly building what she calls \"collaborative identity spaces\" — essentially letting friend groups co-create shared profiles and memory boards. Before starting Mosaic, Vera spent four years at Pinterest on the discovery team, where she became known for her unusually rigorous approach to understanding how people actually browse versus how they say they browse.\n\nWhat makes Vera distinctive in the consumer social space is her blend of analytical depth and genuine collaborative instinct. She's not the type of founder who shows up with a fully-formed vision and expects the team to execute. Instead, she runs these intense working sessions where everyone's hypotheses get stress-tested against user data. Some people find it exhausting; others thrive in it. Her early employees tend to be the latter type.\n\nVera grew up in Vancouver before moving to the Bay Area for undergrad at Stanford, where she studied symbolic systems. She's mentioned in interviews that her interest in social software came from watching her extended family struggle to stay connected across three countries — the tools always felt either too formal or too ephemeral. [Mosaic](companies/mosaic-14) is her attempt to find something in between.\n\nShe's known for being deeply skeptical of vanity metrics. At a YC dinner last year, she apparently got into a heated debate about whether DAU even means anything for a product designed around periodic, meaningful check-ins rather than daily engagement. Her position was that the industry's obsession with daily usage actively corrupts product decisions. Not everyone agreed, but the conversation stuck with people.\n\nVera keeps a relatively low profile on social media, which some find ironic given her work. She's said she prefers to be a user researcher first and a public figure never. Tends to wear the same rotation of three or four sweaters to meetings. Drinks an alarming amount of oolong tea.",
-  "timeline": "- **2021-06-15** | Vera leaves Pinterest after four years on discovery team; announces she's taking time to explore new ideas\n- **2022-01-22** | Incorporates [Mosaic](companies/mosaic-14) in Delaware; starts recruiting founding team\n- **2022-08-03** | Closes $2.4M pre-seed round; moves into small office in Hayes Valley\n- **2023-02-14** | Mosaic launches private beta with ~400 users from Vera's network\n- **2023-09-19** | Gives talk at a small product conference on \"Metrics That Lie\" — gains some attention on Twitter\n- **2024-03-07** | Vera hires former Snap researcher as head of user insights\n- **2024-11-12** | [Mosaic](companies/mosaic-14) opens public waitlist; reaches 12,000 signups in first week\n- **2025-04-28** | Profiled in The Information as one of \"founders rethinking social from scratch\"\n- **2025-08-15** | Announces Mosaic crossing 50k active groups; still hasn't raised Series A",
-  "_facts": {
-    "type": "person",
-    "slug": "people/vera-chen-14",
-    "name": "Vera Chen",
-    "role": "founder",
-    "primary_affiliation": "companies/mosaic-14",
-    "notable_traits": [
-      "collaborative",
-      "analytical"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__vera-rodriguez-171.json b/eval/data/world-v1/people__vera-rodriguez-171.json
deleted file mode 100644
index 5e70644c0..000000000
--- a/eval/data/world-v1/people__vera-rodriguez-171.json
+++ /dev/null
@@ -1,21 +0,0 @@
-{
-  "slug": "people/vera-rodriguez-171",
-  "type": "person",
-  "title": "Vera Rodriguez",
-  "compiled_truth": "Vera Rodriguez is a seasoned climate tech advisor known for her analytical rigor and collaborative approach to supporting early-stage companies. She currently serves as a primary advisor to [Lucid](companies/lucid-21), a climate tech startup working on next-generation carbon capture monitoring systems. Her involvement with Lucid began in late 2022, when the founding team sought her out specifically for her experience navigating complex regulatory environments and building partnerships with industrial players.\n\nBefore transitioning into advisory work, Vera spent nearly a decade in operational roles at various cleantech companies, including a stint as VP of Strategy at a solar infrastructure firm that was later aquired by a major utility. This background gives her a practical edge—she's not just theorizing about market dynamics, she's lived through the messy realities of scaling hardware-intensive businesses. Colleagues describe her as someone who asks the hard questions early, saving teams from costly pivots down the line.\n\nIn addition to her work with Lucid, Rodriguez maintains an advisory relationship with [Umbra](companies/umbra-48), though her engagement there is less intensive. She's been helpful in connecting Umbra's leadership to potential enterprise customers and has participated in several stratgic planning sessions over the past year. Her network across the climate and industrial sectors is extensive, built over years of conference speaking, board participation, and informal mentorship.\n\nVera is based in Denver but travels frequently to the Bay Area and occasionally to Europe for climate summits. She holds an MBA from Stanford and an undergraduate degree in environmental engineering from MIT. Known for being direct but never dismissive, she has a reputation for making founders feel heard even when delivering tough feedback. Her analytical nature means she tends to dig deep into unit economics and technical feasibility before offering strategic guidance.\n\nShe's currently focused on helping Lucid close its Series A and expand its pilot programs with two major cement manufacturers.",
-  "timeline": "- **2021-03-15** | Joined advisory board of early-stage geothermal startup (since wound down)\n- **2022-09-08** | First meeting with [Lucid](companies/lucid-21) founders; agreed to advisory role\n- **2022-11-20** | Helped Lucid secure intro to DOE program officer for grant application\n- **2023-04-12** | Began informal advisory relationship with [Umbra](companies/umbra-48)\n- **2023-08-30** | Spoke on \"Scaling Climate Hardware\" panel at Climate Week NYC\n- **2024-01-17** | Led strategic offsite for Lucid team in Boulder, CO\n- **2024-06-05** | Connected Umbra with potential Series B lead investor\n- **2024-11-02** | Published op-ed in Canary Media on carbon accounting standards\n- **2025-02-19** | Participated in Lucid board meeting; reviewed Series A term sheets\n- **2025-05-10** | Scheduled to keynote at Denver Climate Innovation Summit",
-  "_facts": {
-    "type": "person",
-    "slug": "people/vera-rodriguez-171",
-    "name": "Vera Rodriguez",
-    "role": "advisor",
-    "primary_affiliation": "companies/lucid-21",
-    "secondary_affiliations": [
-      "companies/umbra-48"
-    ],
-    "notable_traits": [
-      "collaborative",
-      "analytical"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__vera-singh-20.json b/eval/data/world-v1/people__vera-singh-20.json
deleted file mode 100644
index 85c69a30f..000000000
--- a/eval/data/world-v1/people__vera-singh-20.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/vera-singh-20",
-  "type": "person",
-  "title": "Vera Singh",
-  "compiled_truth": "Vera Singh is the founder and CEO of [Kindle](companies/kindle-20), a climate tech startup working on next-generation carbon capture infrastructure. Before launching Kindle in 2022, she spent six years at McKinsey where she led sustainability practice engagements across North America and Europe. Her background combines an MBA from Stanford with an undergraduate degree in chemical engineering from IIT Bombay.\n\nWhat makes Vera distinctive in the climate tech space is her ability to bridge hard science and compelling narrative. She's known as a storyteller who can make complex carbon sequestration chemistry accessible to investors, policymakers, and the general public alike. At the same time, Singh brings rigorous analytical chops to the table—former colleagues describe her as someone who won't greenlight a project without seeing the numbers three different ways.\n\nSingh founded [Kindle](companies/kindle-20) after becoming frustrated with the slow pace of corporate sustainability initiatives. She's spoken publicly about a pivotal moment during a client engagement when she realized the gap between announced climate commitments and actual capital deployment. \"Everyone was writing reports about net zero by 2040,\" she said in a 2023 podcast interview. \"Nobody was building the infrastructure to get there.\"\n\nKindle's approach focuses on modular direct air capture units that can be deployed at industrial sites, capturing emissions at the source rather than relying on centralized facilities. The company has raised over $45 million to date and employs roughly 60 people across offices in San Francisco and Mumbai.\n\nVera is also an active angel investor, with a particular interest in climate adaptation technologies and sustainable agriculture. She sits on the advisory board of two university climate labs and frequently speaks at conferences about the intersection of storytelling and climate action. Her TED talk from 2024 on \"Making Carbon Visible\" has been viewed over 2 million times. Colleagues note she's intensly focused but maintains a dry sense of humor that keeps team morale high during difficult stretches.",
-  "timeline": "- **2021-08-15** | Left McKinsey after six years to begin working on climate tech startup concept\n- **2022-03-02** | Officially incorporated [Kindle](companies/kindle-20) in Delaware, began recruiting founding team\n- **2022-09-18** | Closed $8M seed round led by Lowercarbon Capital\n- **2023-04-07** | First pilot unit deployed at cement plant in Gujarat, India\n- **2023-11-12** | Vera delivered keynote at Climate Week NYC on distributed capture infrastructure\n- **2024-02-28** | [Kindle](companies/kindle-20) announced $37M Series A, Vera profiled in Bloomberg Green\n- **2024-06-15** | TED talk \"Making Carbon Visible\" published, quickly went viral\n- **2024-10-03** | Opened Mumbai R&D office, hired 20 engineers\n- **2025-01-22** | Named to Forbes 30 Under 40 Climate Leaders list\n- **2025-08-14** | Announced partnership with three major European industrial firms for 2026 deployments",
-  "_facts": {
-    "type": "person",
-    "slug": "people/vera-singh-20",
-    "name": "Vera Singh",
-    "role": "founder",
-    "primary_affiliation": "companies/kindle-20",
-    "notable_traits": [
-      "storyteller",
-      "analytical"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__vera-wilson-25.json b/eval/data/world-v1/people__vera-wilson-25.json
deleted file mode 100644
index 321070509..000000000
--- a/eval/data/world-v1/people__vera-wilson-25.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/vera-wilson-25",
-  "type": "person",
-  "title": "Vera Wilson",
-  "compiled_truth": "Vera Wilson is the founder and CEO of [Vox](companies/vox-25), an AI applications company that's been making quiet but steady progress in the enterprise space. Known primarily for her exceptional design taste, Vera has built a reputation for shipping products that feel almost unusually polished for an early-stage startup. Colleagues describe her as patient to a fault—willing to delay launches by months if the user experience isn't quite right.\n\nBefore starting Vox, Wilson spent nearly six years at Figma, where she led design systems for their enterprise product line. It was there she developed her thesis that most B2B software fails not because of capability gaps but because of friction and poor information hierarchy. She left in late 2022, taking a few months off before incorporating [Vox](companies/vox-25) in early 2023.\n\nThe company itself focuses on AI-powered workflow tools, though Vera tends to bristle at the \"AI\" label. In interviews she's said that Vox builds \"software that happens to use machine learning\" rather than \"AI products.\" This philosophy shows in how the company markets itself—understated, focused on outcomes rather than technology. Their flagship product helps operations teams automate document processing and routing, apparently saving customers dozens of hours weekly.\n\nVera Wilson's approach to company-building is notably unhurried. She raised a modest seed round and has been deliberate about headcount, keeping the team under twenty people for the first two years. Some investors have questioned whether this patience will cost her market share, but Wilson seems unphased. She's said publicly that she'd rather build something enduring than race to scale prematurely.\n\nOn a personal level, Vera is relatively private. She grew up in Portland, Oregon, studied cognitive science at Berkeley, and briefly considered academia before pivoting to design. She's an amature woodworker and occasionally posts photos of furniture projects. Within the small circle of design-forward founders, she's become something of a cult figure—people reference \"the Vera Wilson standard\" when discussing product quality.",
-  "timeline": "- **2021-03-15** | Promoted to Principal Designer at Figma, leading enterprise design systems team\n- **2022-09-30** | Announced departure from Figma after nearly six years\n- **2023-02-14** | Incorporated [Vox](companies/vox-25) as solo founder\n- **2023-06-22** | Closed $3.2M seed round, lead investor undisclosed\n- **2023-11-08** | Shipped first beta of Vox document routing product to three pilot customers\n- **2024-04-19** | Gave talk at Config 2024 on \"Patience as Product Strategy\"\n- **2024-08-03** | Vox reaches 40 paying enterprise customers\n- **2025-01-27** | Hired first head of engineering, former Stripe infrastructure lead\n- **2025-05-12** | Featured in Fast Company's \"Most Creative People in Business\" list\n- **2026-02-09** | [Vox](companies/vox-25) reportedly in discussions for Series A at $45M valuation",
-  "_facts": {
-    "type": "person",
-    "slug": "people/vera-wilson-25",
-    "name": "Vera Wilson",
-    "role": "founder",
-    "primary_affiliation": "companies/vox-25",
-    "notable_traits": [
-      "design taste",
-      "patient"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__victor-jackson-116.json b/eval/data/world-v1/people__victor-jackson-116.json
deleted file mode 100644
index d004be44c..000000000
--- a/eval/data/world-v1/people__victor-jackson-116.json
+++ /dev/null
@@ -1,19 +0,0 @@
-{
-  "slug": "people/victor-jackson-116",
-  "type": "person",
-  "title": "Victor Jackson",
-  "compiled_truth": "Victor Jackson is a senior engineer at [Vector](companies/vector-6), a health tech company building infrastructure for patient data interoperability. He joined Vector in early 2023 after a stint at a Series B fintech that didn't quite make it through the 2022 downturn. Before that, Victor spent four years at Stripe on their payments platform team, where he developed a reputation for shipping reliably and mentoring junior engineers.\n\nWhat makes Victor unusual for an engineer is his comfort with go-to-market work. He's not the type to hide in the codebase—he'll jump on customer calls, help sales understand technical nuances, and has even run a few demos himself when the team was stretched thin. This GTM-heavy orientation has made him invaluable at [Vector](companies/vector-6), where the sales cycle involves convincing skeptical hospital IT departments that yet another integration won't break their fragile systems.\n\nVictor's recruiting strength is probably his most talked-about trait internally. He's personally responsible for bringing in at least six engineers over the past eighteen months, including two staff-level hires who had competing offers from FAANG companies. His secret seems to be genuine follow-through—he remembers details about candidates' lives, sends thoughtful articles, and makes the pitch feel less like a transaction. The eng hiring pipeline at Vector runs partly on his network and partly on his ability to close.\n\nHe's based in Austin but travels to San Francisco maybe once a month for in-person time with the team. Known to be direct in feedback, sometimes to the point of bluntness. Prefers written communication over meetings. Has a side interest in developer tooling and occasionally contributes to open source projects on weekends, though he's scaled that back since joining Vector. Colleagues describe him as low-ego but high-standards—he doesn't care who gets credit but he does care if the code is sloppy.\n\nVictor studied computer science at Georgia Tech, graduating in 2016. He's 31, married, no kids yet. Runs half-marathons occasionally.",
-  "timeline": "- **2021-03-15** | Promoted to senior engineer at Stripe after leading payments reliability initiative\n- **2022-06-01** | Left Stripe to join Finley (Series B fintech) as eng lead\n- **2022-11-20** | Finley announces layoffs; Victor starts exploring new opportunities\n- **2023-02-14** | Joined [Vector](companies/vector-6) as senior engineer, employee #23\n- **2023-08-09** | Closed hire for staff engineer Maya Chen after 6-week recruiting effort\n- **2024-01-22** | Led technical integration with major hospital network in Texas\n- **2024-05-30** | Gave internal talk on 'Engineers in the Sales Process' that became company lore\n- **2024-09-12** | Recruited two engineers from former Stripe team to join Vector\n- **2025-02-03** | Promoted to tech lead for the integrations team\n- **2025-04-18** | Participated in panel at Health Tech Connect conference in Boston",
-  "_facts": {
-    "type": "person",
-    "slug": "people/victor-jackson-116",
-    "name": "Victor Jackson",
-    "role": "engineer",
-    "primary_affiliation": "companies/vector-6",
-    "secondary_affiliations": [],
-    "notable_traits": [
-      "recruiting strength",
-      "GTM-heavy"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__victor-taylor-1.json b/eval/data/world-v1/people__victor-taylor-1.json
deleted file mode 100644
index c1210d5c1..000000000
--- a/eval/data/world-v1/people__victor-taylor-1.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/victor-taylor-1",
-  "type": "person",
-  "title": "Victor Taylor",
-  "compiled_truth": "Victor Taylor is the founder of [Beta](companies/beta-1), a cybersecurity company that's built a reputation for catching what others miss. Before starting Beta, Victor spent nearly a decade bouncing between threat intelligence roles at various defense contractors and a brief stint at a major cloud provider where he led incident response. He's known in security circles as a sharp pattern matcher—the kind of person who can look at disparate log data and intuit the attack vector before the tooling catches up.\n\nVictor's approach to building [Beta](companies/beta-1) reflects his first-principles thinking. Rather than layering more detection rules onto existing frameworks, he pushed his team to rethink what endpoint security should look like from scratch. This led to Beta's core product: a lightweight agent that relies heavily on behavioral analysis rather than signature matching. The bet was controversial early on, but it's paid off as adversaries have gotten better at evading traditional defenses.\n\nThose who've worked with Taylor describe him as intense but fair. He has a habit of asking \"why\" until he hits bedrock, which can be exhausting in meetings but tends to surface assumptions nobody else was questioning. He's not a polished public speaker—tends to ramble a bit when he gets excited about a technical tangent—but his conference talks consistently draw crowds because the content is genuinely novel.\n\nVictor grew up in the Pacific Northwest and studied computer science at University of Washington, though he dropped out his senior year to join a startup that was later aquired by Symantec. He doesn't talk much about the Symantec years, but collegues from that era say it shaped his views on what not to do when scaling a security company. He's been quoted saying that most enterprise security software is \"designed to check compliance boxes, not stop attackers.\"\n\nHe lives in Seattle with his wife and two kids. Outside of work, he's an avid rock climber and has been known to disappear for weeks at a time to remote climbing destinations when Beta hits major milestones.",
-  "timeline": "- **2021-03-15** | Victor officially incorporates [Beta](companies/beta-1) after months of prototyping the core detection engine\n- **2021-09-02** | First paying customer signed—a mid-sized fintech that had been breached twice in the prior year\n- **2022-04-18** | Keynote at BSides Seattle on behavioral detection methods; video later circulates widely on security Twitter\n- **2022-11-30** | Beta closes seed round; Victor insists on keeping the round small to maintain control\n- **2023-06-12** | Hires first VP of Engineering after months of failed searches\n- **2023-10-07** | Victor presents at a private CISO roundtable in San Francisco, generates significant enterprise pipeline\n- **2024-02-22** | Beta detects novel supply chain attack affecting three Fortune 500 companies; Victor handles disclosure personally\n- **2024-08-14** | Featured in Wired profile on \"founders rethinking cybersecurity\"\n- **2025-01-09** | Announces Series A on LinkedIn with characteristically understated post\n- **2025-11-03** | Victor takes a three-week climbing trip to Patagonia after Beta crosses 500 enterprise customers",
-  "_facts": {
-    "type": "person",
-    "slug": "people/victor-taylor-1",
-    "name": "Victor Taylor",
-    "role": "founder",
-    "primary_affiliation": "companies/beta-1",
-    "notable_traits": [
-      "sharp pattern matcher",
-      "first-principles thinker"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__victor-wilson-3.json b/eval/data/world-v1/people__victor-wilson-3.json
deleted file mode 100644
index faf4bc808..000000000
--- a/eval/data/world-v1/people__victor-wilson-3.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/victor-wilson-3",
-  "type": "person",
-  "title": "Victor Wilson",
-  "compiled_truth": "Victor Wilson is the founder and driving force behind [Delta](companies/delta-3), a biotech startup that's been making waves in synthetic biology infrastructure. Known for his relentless pace and collaborative approach, Victor has built a reputation as someone who ships fast and iterates faster—a rare combination in an industry often bogged down by regulatory caution and academic perfectionism.\n\nBefore launching Delta, Victor spent nearly a decade in computational biology, first at a major pharma company and then at a smaller genomics outfit where he led platform development. His frustration with slow-moving institutional science eventually pushed him to strike out on his own. He's talked openly about how the traditional biotech playbook felt broken to him—too much time spent in stealth mode, not enough real-world feedback loops.\n\nAt [Delta](companies/delta-3), Wilson has championed an unusually open development process. The company shares internal roadmaps publicly and actively solicits input from potential customers and academic partners during early R&D phases. This collabroative ethos has attracted a tight-knit team of engineers and biologists who share his impatience with convention. Victor often says that \"shipping is the only real validation\" and has been known to push prototype tools into beta within weeks of initial concept.\n\nHis leadership style is hands-on but not micromanaging. Team members describe him as intensely curious, always asking questions and digging into technical details, but ultimately trusting his people to execute. He's also built a network of advisors and collaborators across the biotech ecosystem, frequently co-authoring white papers and appearing on panels alongside researchers and other founders.\n\nVictor Wilson's background includes a PhD in systems biology from Stanford and a brief stint doing postdoctoral work before pivoting to industry. He's spoken at multiple synthetic biology conferences and was featred in a recent profile by a major science publication highlighting new approaches to biotech entrepreneurship. Despite Delta's rapid growth, he's maintained a relatively low public profile, preferring to let the company's output speak for itself.",
-  "timeline": "- **2021-03-15** | Victor Wilson incorporates [Delta](companies/delta-3) after leaving his previous role in genomics platform development\n- **2021-09-02** | Presents early Delta prototype at SynBioBeta conference, generates significant interest from potential partners\n- **2022-01-18** | Closes seed round, brings on first full-time engineering hire\n- **2022-07-11** | Delta ships first public beta of its core synthetic biology toolkit\n- **2023-02-25** | Co-authors open white paper on modular biotech infrastructure with academic collaborators\n- **2023-10-09** | Featured in Wired profile on fast-moving biotech founders\n- **2024-04-14** | Announces strategic partnership with two university research labs for pilot programs\n- **2024-11-30** | Victor speaks at Founders in Bio summit on collaborative R&D models\n- **2025-06-22** | [Delta](companies/delta-3) reaches 50 employees; Wilson promotes two co-leads to expand leadership team",
-  "_facts": {
-    "type": "person",
-    "slug": "people/victor-wilson-3",
-    "name": "Victor Wilson",
-    "role": "founder",
-    "primary_affiliation": "companies/delta-3",
-    "notable_traits": [
-      "fast-shipping",
-      "collaborative"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__wendy-hernandez-80.json b/eval/data/world-v1/people__wendy-hernandez-80.json
deleted file mode 100644
index fb6687545..000000000
--- a/eval/data/world-v1/people__wendy-hernandez-80.json
+++ /dev/null
@@ -1,25 +0,0 @@
-{
-  "slug": "people/wendy-hernandez-80",
-  "type": "person",
-  "title": "Wendy Hernandez",
-  "compiled_truth": "Wendy Hernandez is a partner at [Founders Fund](companies/founders-fund-0), one of the most influential venture capital firms in Silicon Valley. Known for her patient approach to company building and an almost uncanny ability to identify and recruit top-tier talent, she's become a go-to board member for founders who prioritize long-term vision over quick exits.\n\nBefore joining Founders Fund, Wendy spent nearly a decade in operational roles at early-stage startups, giving her a grounded perspective that many pure-finance VCs lack. She understands the trenches—the missed payrolls, the pivot conversations, the 2am product debates. This background makes her particuarly effective when advising portfolio companies through rough patches.\n\nHer current board seats reflect a diverse but thematically connected portfolio. She sits on the boards of [Cipher Labs](companies/cipher-labs-63), a cryptography-focused infrastructure company, and [Drift](companies/drift-31), which is building next-gen communication tools for distributed teams. She's also involved with [Spire](companies/spire-46), [Delta Labs](companies/delta-labs-53), and [Vellum Labs](companies/vellum-labs-99)—all companies operating at the intersection of developer tools and applied AI.\n\nHernandez is perhaps best known for her recruiting strength. Multiple founders have credited her with helping land their first five hires, often pulling from her extensive network of operators who trust her judgement implicitly. She tends to focus less on pedigree and more on raw capability and cultural fit, which has led to some unconventional but highly successful team compositions.\n\nHer investment style is decidedly patient. Wendy has publicly stated that she's comfortable holding positions for 10+ years if the underlying thesis remains intact. This long-term orientation has occasionally put her at odds with more aggressive partners, but her track record speaks for itself. She led the Series A for Cipher Labs back in 2022 when most investors were skeptical of pure cryptography plays, and that bet is looking increasingly prescient.\n\nOutside of work, Wendy is relatively private. She speaks at conferences sparingly, preferring small dinners and one-on-one founder meetings. When she does appear publicly, she's known for blunt, actionable advice rather than platitudes.",
-  "timeline": "- **2021-03-15** | Joined [Founders Fund](companies/founders-fund-0) as a partner after being recruited by the leadership team\n- **2022-01-22** | Led Series A investment in [Cipher Labs](companies/cipher-labs-63), her first major solo deal at the firm\n- **2022-09-08** | Joined the board of [Drift](companies/drift-31) following their seed extension round\n- **2023-04-11** | Spoke at a private founder dinner in SF about patience in venture and the tyranny of fast markups\n- **2023-11-30** | Took board observer seat at [Spire](companies/spire-46) as part of bridge financing\n- **2024-02-14** | Helped recruit CTO for [Delta Labs](companies/delta-labs-53), pulling from her network at Stripe alumni\n- **2024-08-19** | Participated in [Vellum Labs](companies/vellum-labs-99) Series B alongside existing investors\n- **2025-01-07** | Internal Founders Fund memo leaked praising Hernandez's portfolio performance through 2024\n- **2025-06-22** | Hosted intimate LP dinner discussing AI infrastructure thesis and long-term holds",
-  "_facts": {
-    "type": "person",
-    "slug": "people/wendy-hernandez-80",
-    "name": "Wendy Hernandez",
-    "role": "partner",
-    "primary_affiliation": "companies/founders-fund-0",
-    "secondary_affiliations": [
-      "companies/cipher-labs-63",
-      "companies/drift-31",
-      "companies/spire-46",
-      "companies/delta-labs-53",
-      "companies/vellum-labs-99"
-    ],
-    "notable_traits": [
-      "recruiting strength",
-      "patient"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__wendy-wilson-170.json b/eval/data/world-v1/people__wendy-wilson-170.json
deleted file mode 100644
index ff69d202a..000000000
--- a/eval/data/world-v1/people__wendy-wilson-170.json
+++ /dev/null
@@ -1,23 +0,0 @@
-{
-  "slug": "people/wendy-wilson-170",
-  "type": "person",
-  "title": "Wendy Wilson",
-  "compiled_truth": "Wendy Wilson is a seasoned technical advisor known for her patient approach and deep expertise across climate technology, infrastructure, and applied research. She currently serves as an advisor to [Delta Labs](companies/delta-labs-53), a climate tech company working on atmospheric carbon capture and monitoring systems. Her involvement with Delta Labs began in late 2022 when the company was pivoting from pure research to commercial applications—a transition where her steady hand proved invaluable.\n\nBefore taking on advisory roles full-time, Wilson spent over a decade in engineering leadership at various startups and research institutions. She's the type of advisor who actually reads the technical documentation, often catching edge cases that founding teams miss in their rush to ship. This technical depth, combined with unusual patience for early-stage chaos, has made her a sought-after voice in founder circles.\n\nWendy also advises [Vector](companies/vector-6), where she focuses on their data infrastructure stack, and maintains board observer seats at both [Lattice Labs](companies/lattice-labs-89) and [Acme Labs](companies/acme-labs-50). Her cross-pollination across these companies creates interesting knowledge transfer—she's been known to connect founders facing similar scaling challenges, though she's careful about confidentiality boundries.\n\nThose who work with Wilson often note her distinctive communication style. She asks more questions than she answers, at least initially. \"Wendy doesn't tell you what to do,\" one founder remarked. \"She helps you realize what you already knew but were afraid to admit.\" This socratic method can frustrate founders looking for quick answers, but tends to produce more durable decisions.\n\nShe holds a PhD in materials science from MIT and did postdoctoral work at Lawrence Berkeley National Laboratory. Her academic background informs her advisory work—she's particularly effective with deep tech companies navigating the valley of death between lab results and production. Wilson splits her time between the Bay Area and Boulder, where she ocasionally guest lectures at CU's engineering school.",
-  "timeline": "- **2021-03-15** | Joined [Acme Labs](companies/acme-labs-50) as technical advisor following introduction from a mutual investor\n- **2021-09-22** | Keynote talk at Climate Tech Summit on commercializing atmospheric research\n- **2022-04-08** | Began advising [Lattice Labs](companies/lattice-labs-89) on their sensor calibration methodology\n- **2022-11-30** | Started formal advisory role at [Delta Labs](companies/delta-labs-53) during their Series A process\n- **2023-06-14** | Helped [Vector](companies/vector-6) navigate critical infrastructure decision ahead of enterprise launch\n- **2023-12-01** | Published essay on patience in deep tech building, widely circulated in founder communities\n- **2024-05-19** | Facilitated introduction between Delta Labs and potential strategic partner\n- **2024-10-07** | Advisory agreement with [Delta Labs](companies/delta-labs-53) extended through 2026\n- **2025-02-22** | Guest lecture at CU Boulder on climate tech commercialization pathways\n- **2025-08-11** | Participated in [Lattice Labs](companies/lattice-labs-89) board strategy session on international expansion",
-  "_facts": {
-    "type": "person",
-    "slug": "people/wendy-wilson-170",
-    "name": "Wendy Wilson",
-    "role": "advisor",
-    "primary_affiliation": "companies/delta-labs-53",
-    "secondary_affiliations": [
-      "companies/vector-6",
-      "companies/lattice-labs-89",
-      "companies/acme-labs-50"
-    ],
-    "notable_traits": [
-      "patient",
-      "technical depth"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__xavier-nakamura-118.json b/eval/data/world-v1/people__xavier-nakamura-118.json
deleted file mode 100644
index a610f6b43..000000000
--- a/eval/data/world-v1/people__xavier-nakamura-118.json
+++ /dev/null
@@ -1,21 +0,0 @@
-{
-  "slug": "people/xavier-nakamura-118",
-  "type": "person",
-  "title": "Xavier Nakamura",
-  "compiled_truth": "Xavier Nakamura is a senior engineer at [Pulse](companies/pulse-8), an edtech startup focused on adaptive learning platforms for K-12 students. Known for his fast-shipping mentality, Xavier has become something of a legend internally for pushing features from concept to production in timeframes that make other engineers nervous. He joined Pulse in early 2022 after a stint at [Vox](companies/vox-25), where he worked on backend infrastructure for their podcast distribution platform.\n\nBefore Vox, Nakamura spent three years at a fintech startup that ultimately folded, which he credits with teaching him how to move fast without breaking things—or at least, how to break things in ways that are easy to fix. He's a first-principles thinker, the kind of person who will question why a system exists before agreeing to optimize it. This has occasionally put him at odds with product managers who just want the feature built, but his track record speaks for itself.\n\nXavier holds a BS in Computer Science from UC San Diego, though he's quick to point out that most of what he actually uses day-to-day he learned on the job or from late-night documentation binges. He's particulary interested in real-time systems and has given a few talks at local meetups on websocket architectures and the tradeoffs between different pub/sub patterns.\n\nAt Pulse, he's been instrumental in building out the adaptive quiz engine that adjusts difficulty based on student performance in real-time. The system processes millions of interactions daily and has become one of the company's key differentiators in the crowded edtech space. Colleagues describe him as intense but approachable—the kind of engineer who will absolutely tell you your idea won't scale, but will also stay late to help you figure out what will.\n\nNakamura maintains a minimal online presence, preferring to let his work speak for itself. He lives in Oakland with two cats named after sorting algorithms.",
-  "timeline": "- **2021-03-15** | Joined [Vox](companies/vox-25) as a backend engineer working on podcast infrastructure\n- **2022-01-10** | Left Vox to join [Pulse](companies/pulse-8) as senior engineer\n- **2022-06-22** | Shipped first version of adaptive quiz engine in under 8 weeks\n- **2023-02-14** | Gave talk at SF Engneering Meetup on real-time system design patterns\n- **2023-09-05** | Promoted to tech lead for Pulse's core learning platform team\n- **2024-03-18** | Led architecture overhaul that reduced quiz latency by 60%\n- **2024-11-02** | Mentioned in TechCrunch piece on edtech infrastructure innovation\n- **2025-04-25** | Started mentoring junior engineers through internal Pulse program\n- **2025-08-30** | Began advising [Vox](companies/vox-25) on scaling challenges as informal consultant",
-  "_facts": {
-    "type": "person",
-    "slug": "people/xavier-nakamura-118",
-    "name": "Xavier Nakamura",
-    "role": "engineer",
-    "primary_affiliation": "companies/pulse-8",
-    "secondary_affiliations": [
-      "companies/vox-25"
-    ],
-    "notable_traits": [
-      "fast-shipping",
-      "first-principles thinker"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__yara-johnson-8.json b/eval/data/world-v1/people__yara-johnson-8.json
deleted file mode 100644
index dd7af9498..000000000
--- a/eval/data/world-v1/people__yara-johnson-8.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/yara-johnson-8",
-  "type": "person",
-  "title": "Yara Johnson",
-  "compiled_truth": "Yara Johnson is the founder and CEO of [Pulse](companies/pulse-8), an edtech company focused on real-time learning analytics and student engagement tools. She's known in founder circles as a first-principles thinker who obsesses over product details to an almost unreasonable degree—the kind of person who will spend three hours debating button placement while her team waits to ship.\n\nBefore starting Pulse, Yara spent four years at a large education nonprofit where she ran product for their digital learning intiatives. The experience left her frustrated with how slow institutions moved and how little they actually understood about student behavior. She quit in 2021 to build what she wished had existed: lightweight instrumentation that helps teachers see who's engaged and who's falling behind, without adding more work to anyone's plate.\n\nYara grew up in Houston, studied cognitive science at Rice, and briefly considered a PhD before deciding academia wasn't for her. She's talked openly about being a first-generation college student and how that shapes her approach to edtech—she's skeptical of solutions that assume students have resources or support systems they often don't.\n\nHer product philosophy is simple but rigid: if a feature doesn't obviously help a teacher or student in the first thirty seconds, it doesn't ship. This has made [Pulse](companies/pulse-8) slower to expand its feature set than competitors, but retention numbers are strong. Teachers who adopt it tend to stick around.\n\nYara is not particularly active on social media, prefering to spend her time talking directly to users. She does a monthly \"office hours\" call with educators that often runs two hours over schedule. People describe her as intense but genuinely curious—she asks a lot of questions and remembers the answers months later. She's been invited to speak at several edtech conferences but has turned most down, saying she'd rather be building than talking about building.",
-  "timeline": "- **2021-03-15** | Yara leaves her role at Education Forward nonprofit to start working on Pulse full-time\n- **2021-08-22** | Incorporates [Pulse](companies/pulse-8) as a Delaware C-corp, begins initial prototype development\n- **2022-01-10** | Closes $1.2M pre-seed round led by Reach Capital\n- **2022-09-04** | Pulse launches beta with 12 pilot schools in Texas\n- **2023-02-18** | Speaks at ASU+GSV Summit on \"Why Most Edtech Analytics Fail Teachers\"\n- **2023-07-30** | [Pulse](companies/pulse-8) hits 200 schools on the platform, mostly organic growth\n- **2024-01-14** | Raises $8M Series A, announces expansion to community colleges\n- **2024-11-02** | Featured in EdSurge profile on founder-led product companies\n- **2025-04-20** | Hires first VP of Engineering after running lean team for three years\n- **2025-09-08** | Announces Pulse integration with major LMS providers",
-  "_facts": {
-    "type": "person",
-    "slug": "people/yara-johnson-8",
-    "name": "Yara Johnson",
-    "role": "founder",
-    "primary_affiliation": "companies/pulse-8",
-    "notable_traits": [
-      "first-principles thinker",
-      "product-obsessed"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__yara-moore-174.json b/eval/data/world-v1/people__yara-moore-174.json
deleted file mode 100644
index e1443cd07..000000000
--- a/eval/data/world-v1/people__yara-moore-174.json
+++ /dev/null
@@ -1,22 +0,0 @@
-{
-  "slug": "people/yara-moore-174",
-  "type": "person",
-  "title": "Yara Moore",
-  "compiled_truth": "Yara Moore is an advisor and operator known for her ability to ship product fast and build high-performing teams from scratch. She currently serves as an advisor to [Tessera Labs](companies/tessera-labs-65), an edtech company working on adaptive learning infrastructure. Her involvement there has coincided with a notable acceleration in their product cadence—three major releases in under eight months.\n\nBefore settling into advisory roles, Yara spent nearly a decade in operating positions across growth-stage startups. She's particuarly known for her recruiting instincts, having built out engineering and go-to-market teams at two companies that later exited. Friends describe her as someone who \"just knows\" when a candidate will work out, though she insists it's mostly about reference checks and pattern matching.\n\nMoore also advises [Vector Labs](companies/vector-labs-56), where she's helped shape their hiring process for senior engineers, and [Tempo](companies/tempo-24), a productivity tools company where she occasionally weighs in on product strategy. Her involvment at Tempo is more hands-off—mostly async feedback and quarterly check-ins—but the founders credit her with pushing them toward a more focused roadmap.\n\nYara tends to avoid the spotlight. She rarely tweets, doesn't maintain a personal blog, and turns down most podcast invitations. When she does speak publicly, it's usually at small founder dinners or private Slack groups. Her advice tends to be practical rather than philosophical: ship something, talk to users, hire slow.\n\nShe lives in Austin but travels frequently to San Francisco and New York for board meetings and founder sessions. Those who've worked with her say she's direct to the point of bluntness, but deeply generous with her time for people she believes in. One founder described her feedback style as \"a punch followed by a hug.\" Yara Moore remains one of the more effective behind-the-scenes operators in the current startup ecosystem, even if few outside it know her name.",
-  "timeline": "- **2021-03-15** | Joined [Tessera Labs](companies/tessera-labs-65) as an advisor following intro from a mutual investor\n- **2021-09-02** | Helped Tessera close their first two senior engineering hires\n- **2022-01-20** | Started advising [Vector Labs](companies/vector-labs-56) on recruiting processes\n- **2022-07-11** | Spoke at a private founder dinner in SF on building early teams\n- **2023-02-08** | Began light advisory relationship with [Tempo](companies/tempo-24)\n- **2023-06-19** | Tessera ships adaptive assessment engine; Yara credited with pushing timeline\n- **2024-01-25** | Facilitated key intro between Tessera and a strategic distribution partner\n- **2024-08-30** | Vector Labs closes Series A; Yara involved in exec recruiting during round\n- **2025-04-12** | Participated in internal Tempo offsite on 2025 product roadmap\n- **2025-11-03** | Named to a private list of \"most effective startup advisors\" circulated among VCs",
-  "_facts": {
-    "type": "person",
-    "slug": "people/yara-moore-174",
-    "name": "Yara Moore",
-    "role": "advisor",
-    "primary_affiliation": "companies/tessera-labs-65",
-    "secondary_affiliations": [
-      "companies/vector-labs-56",
-      "companies/tempo-24"
-    ],
-    "notable_traits": [
-      "fast-shipping",
-      "recruiting strength"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/data/world-v1/people__yara-smith-30.json b/eval/data/world-v1/people__yara-smith-30.json
deleted file mode 100644
index a5ebdf9a3..000000000
--- a/eval/data/world-v1/people__yara-smith-30.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
-  "slug": "people/yara-smith-30",
-  "type": "person",
-  "title": "Yara Smith",
-  "compiled_truth": "Yara Smith is the founder of [Cascade](companies/cascade-30), an AI applications company that's been making waves in the enterprise automation space. She's known primarily for two things: an almost supernatural ability to recruit top-tier talent, and a technical depth that surprises people who expect founders to be pure business types.\n\nBefore starting Cascade, Yara spent six years at Google working on internal ML tooling, eventually leading a team of 40 engineers. She left in 2022, reportedly frustrated with the pace of shipping. Those who worked with her describe someone who could context-switch between architecture discussions and candidate closing calls without missing a beat. One former colleague noted she \"interviews like a sniper\" — identifying exactly what motivates candidates and constructing offers they can't refuse.\n\nThe technical chops are real though. Smith still reviews PRs weekly and has been known to jump into the codebase during crunch periods. Her background is in distributed systems, which shows in how Cascade's platform handles scale. She's published a few papers on fault-tolerant ML pipelines, nothing groundbreaking but solid work that demonstrates she understands the stack deeply.\n\nYara's recruiting strength has been central to [Cascade](companies/cascade-30)'s early success. In under two years she assembled a team that includes three ex-DeepMind researchers and a former Stripe infrastructure lead. People who've turned down offers elsewhere somehow end up at Cascade. Part of it is her pitch — she's compelling on the vision — but insiders say she's also just relentless in follow-up. Candidates get personalized notes, intros to team members, whatever it takes.\n\nShe's relatively low-profile publicly, preferring to let the product speak. Does the occasional podcast but avoids the conference circuit. Lives in San Francisco, has mentioned in interviews she's a morning person who codes before 7am. Married, one kid. The bio stuff rarely comes up — she keeps work and personal fairly seperate.\n\nAt 34, she's younger than most enterprise founders but carries herself with a confidence that reads as earned rather than performed.",
-  "timeline": "- **2022-03-15** | Yara Smith leaves Google after six years, announces she's starting something new\n- **2022-06-01** | Incorporates [Cascade](companies/cascade-30), begins recruiting founding team\n- **2022-09-22** | Closes $4M seed round, mostly from angels and former colleagues\n- **2023-02-10** | Hires Maria Chen (ex-DeepMind) as head of research\n- **2023-07-18** | Cascade launches private beta with three enterprise customers\n- **2023-11-30** | Featured in The Information piece on \"technical founders to watch\"\n- **2024-04-05** | Series A announced, $18M led by Index Ventures\n- **2024-08-12** | Gives rare public talk at AI Eng Summit on building reliable AI systems\n- **2025-01-20** | Cascade hits 50 employees, mostly engineers\n- **2025-06-03** | Spotted meeting with Anthropic leadership, sparking partnership rumors",
-  "_facts": {
-    "type": "person",
-    "slug": "people/yara-smith-30",
-    "name": "Yara Smith",
-    "role": "founder",
-    "primary_affiliation": "companies/cascade-30",
-    "notable_traits": [
-      "recruiting strength",
-      "technical depth"
-    ]
-  }
-}
\ No newline at end of file
diff --git a/eval/generators/gen.ts b/eval/generators/gen.ts
deleted file mode 100644
index 3aaa900a6..000000000
--- a/eval/generators/gen.ts
+++ /dev/null
@@ -1,292 +0,0 @@
-/**
- * Opus prose generator. Calls Claude Opus to turn entity skeletons (from
- * world.ts) into rich, multi-paragraph prose pages with realistic noise:
- * varied phrasing, occasional typos, multiple mentions per page, evolving
- * compiled truth.
- *
- * Cost discipline:
- *   - Tracks token usage per call.
- *   - Hard-stops at $80 (well under the $500 daily cap).
- *   - Caches every successful output to eval/data/world-v1/<slug>.json so
- *     re-running is free.
- *
- * Reads ANTHROPIC_API_KEY from .env.testing (gitignored — committed key
- * would be a security issue).
- *
- * Usage: bun eval/generators/gen.ts [--max N] [--dry-run]
- */
-
-import Anthropic from '@anthropic-ai/sdk';
-import { readFileSync, writeFileSync, existsSync, mkdirSync, readdirSync } from 'fs';
-import { join } from 'path';
-import { buildWorld, type EntityFacts, type World } from './world.ts';
-
-// ─── Setup: load env, init client ──────────────────────────────
-
-function loadEnv() {
-  const envPath = '.env.testing';
-  if (!existsSync(envPath)) throw new Error(`${envPath} not found`);
-  const content = readFileSync(envPath, 'utf-8');
-  for (const line of content.split('\n')) {
-    const m = line.match(/^([A-Z_]+)=(.*)$/);
-    if (m) process.env[m[1]] = m[2].replace(/^["']|["']$/g, '');
-  }
-}
-
-loadEnv();
-if (!process.env.ANTHROPIC_API_KEY) {
-  console.error('ANTHROPIC_API_KEY not set after loading .env.testing');
-  process.exit(1);
-}
-
-const client = new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY });
-
-// Opus 4.7 pricing as of 2026-04-18.
-const PRICE_INPUT_PER_M = 15;
-const PRICE_OUTPUT_PER_M = 75;
-const HARD_STOP_USD = 80;
-const MODEL = 'claude-opus-4-5'; // 4.7 model id; SDK accepts the alias
-
-// ─── Prompt construction ──────────────────────────────────────
-
-function entityPrompt(entity: EntityFacts, world: World): string {
-  // Build a context block that names the related entities so the LLM uses real slugs.
-  let context = '';
-  if (entity.type === 'person') {
-    const company = world.companies.find(c => c.slug === entity.primary_affiliation);
-    const secondaryCos = (entity.secondary_affiliations ?? [])
-      .map(s => world.companies.find(c => c.slug === s))
-      .filter(Boolean);
-    context = `Person profile to write about:
-  Name: ${entity.name}
-  Slug: ${entity.slug}
-  Role: ${entity.role}
-  Primary affiliation: [${company?.name ?? '?'}](${entity.primary_affiliation}) (${company?.industry ?? '?'})
-${secondaryCos.length ? `  Other affiliations: ${secondaryCos.map(c => `[${c!.name}](${c!.slug})`).join(', ')}` : ''}
-  Notable traits: ${entity.notable_traits.join(', ')}`;
-  } else if (entity.type === 'company') {
-    const founders = (entity.founders ?? []).map(s => world.people.find(p => p.slug === s)).filter(Boolean);
-    const investors = (entity.investors ?? []).slice(0, 3).map(s => world.people.find(p => p.slug === s)).filter(Boolean);
-    const advisors = (entity.advisors ?? []).slice(0, 2).map(s => world.people.find(p => p.slug === s)).filter(Boolean);
-    context = `Company profile to write about:
-  Name: ${entity.name}
-  Slug: ${entity.slug}
-  Category: ${entity.category}
-  Industry: ${entity.industry}
-${entity.founded_year ? `  Founded: ${entity.founded_year}` : ''}
-${founders.length ? `  Founders: ${founders.map(p => `[${p!.name}](${p!.slug})`).join(', ')}` : ''}
-${investors.length ? `  Investors: ${investors.map(p => `[${p!.name}](${p!.slug})`).join(', ')}` : ''}
-${advisors.length ? `  Advisors: ${advisors.map(p => `[${p!.name}](${p!.slug})`).join(', ')}` : ''}`;
-  } else if (entity.type === 'meeting') {
-    const attendees = entity.attendees.map(s => world.people.find(p => p.slug === s)).filter(Boolean);
-    const company = entity.topic_company ? world.companies.find(c => c.slug === entity.topic_company) : null;
-    context = `Meeting to write notes for:
-  Name: ${entity.name}
-  Slug: ${entity.slug}
-  Type: ${entity.meeting_type}
-  Date: ${entity.date}
-  Attendees: ${attendees.map(p => `[${p!.name}](${p!.slug})`).join(', ')}
-${company ? `  Topic company: [${company.name}](${company.slug}) (${company.industry})` : ''}`;
-  } else {
-    const cos = entity.related_companies.map(s => world.companies.find(c => c.slug === s)).filter(Boolean);
-    context = `Concept to write a thesis page for:
-  Name: ${entity.name}
-  Slug: ${entity.slug}
-  Brief: ${entity.description}
-  Related companies: ${cos.map(c => `[${c!.name}](${c!.slug})`).join(', ')}`;
-  }
-
-  return `${context}
-
-Write a brain page for this entity. Output JSON with this exact shape:
-{
-  "title": "Display title for the page",
-  "compiled_truth": "Multi-paragraph current understanding. 250-500 words. NATURAL prose, not bullet lists. Reference other entities by markdown link [Name](slug) at least twice using the slugs given above. Vary writing style — sometimes terse, sometimes prose-heavy. Include a couple of natural typos (1-2% of words). Mention the entity by varying names (full name, short name, role). For companies, include details on what they do, recent moves, who's involved. For people, write a bio that mentions their company, history, what they're known for. For meetings, write attendee notes + key discussion points. For concepts, write a thesis with examples.",
-  "timeline": "5-10 dated bullet entries in the format: - **YYYY-MM-DD** | summary text. Mix of dates spanning 2021-2026. Realistic events: hires, raises, ships, talks, meetings. Reference other entities by [Name](slug) where natural."
-}
-
-Output ONLY the JSON object. No preamble, no code fences.`;
-}
-
-// ─── Cost tracking ───────────────────────────────────────────
-
-interface CostLedger {
-  inputTokens: number;
-  outputTokens: number;
-  costUsd: number;
-  calls: number;
-}
-
-const ledger: CostLedger = { inputTokens: 0, outputTokens: 0, costUsd: 0, calls: 0 };
-
-function recordUsage(inT: number, outT: number) {
-  ledger.inputTokens += inT;
-  ledger.outputTokens += outT;
-  ledger.costUsd = (ledger.inputTokens / 1_000_000) * PRICE_INPUT_PER_M + (ledger.outputTokens / 1_000_000) * PRICE_OUTPUT_PER_M;
-  ledger.calls++;
-}
-
-// ─── Main loop ────────────────────────────────────────────────
-
-const OUTPUT_DIR = 'eval/data/world-v1';
-
-async function generateOne(entity: EntityFacts, world: World): Promise<{ ok: true; cached: boolean } | { ok: false; error: string }> {
-  const cachePath = join(OUTPUT_DIR, `${entity.slug.replace('/', '__')}.json`);
-  if (existsSync(cachePath)) return { ok: true, cached: true };
-
-  if (ledger.costUsd > HARD_STOP_USD) {
-    return { ok: false, error: `HARD_STOP: cost ${ledger.costUsd.toFixed(2)} > ${HARD_STOP_USD}` };
-  }
-
-  const prompt = entityPrompt(entity, world);
-  try {
-    const resp = await client.messages.create({
-      model: MODEL,
-      max_tokens: 2500,
-      messages: [{ role: 'user', content: prompt }],
-    });
-    recordUsage(resp.usage.input_tokens, resp.usage.output_tokens);
-
-    const text = resp.content[0].type === 'text' ? resp.content[0].text : '';
-    // Be lenient with trailing junk — find first { and last }.
-    const start = text.indexOf('{');
-    const end = text.lastIndexOf('}');
-    if (start === -1 || end === -1) return { ok: false, error: `no JSON in response (${text.slice(0, 80)})` };
-    const json = JSON.parse(text.slice(start, end + 1));
-
-    writeFileSync(cachePath, JSON.stringify({
-      slug: entity.slug,
-      type: entity.type,
-      title: json.title,
-      compiled_truth: json.compiled_truth,
-      timeline: json.timeline,
-      _facts: entity, // ground truth for benchmark scoring
-    }, null, 2));
-
-    return { ok: true, cached: false };
-  } catch (e) {
-    const msg = e instanceof Error ? e.message : String(e);
-    return { ok: false, error: msg.slice(0, 200) };
-  }
-}
-
-async function main() {
-  const args = process.argv.slice(2);
-  const maxIdx = args.indexOf('--max');
-  const max = maxIdx !== -1 ? Number(args[maxIdx + 1]) : 240;
-  const concIdx = args.indexOf('--concurrency');
-  const concurrency = concIdx !== -1 ? Number(args[concIdx + 1]) : 1;
-  const dryRun = args.includes('--dry-run');
-
-  if (!existsSync(OUTPUT_DIR)) mkdirSync(OUTPUT_DIR, { recursive: true });
-
-  const world = buildWorld(42);
-
-  // Subset selection: aim for diversity. Take a stratified sample so we have
-  // a mix of types in the prose corpus.
-  // 80 people (40 founders, 20 partners, 10 engineers, 10 advisors)
-  // 80 companies (60 startups, 15 VCs, 5 acquirers)
-  // 50 meetings (15 demo days, 25 oneonones, 10 board meetings)
-  // 30 concepts
-  const founders = world.people.filter(p => p.role === 'founder');
-  const partners = world.people.filter(p => p.role === 'partner');
-  const engineers = world.people.filter(p => p.role === 'engineer');
-  const advisors = world.people.filter(p => p.role === 'advisor');
-  const startups = world.companies.filter(c => c.category === 'startup');
-  const vcs = world.companies.filter(c => c.category === 'vc');
-  const acquirers = world.companies.filter(c => c.category === 'acquirer' || c.category === 'mature');
-  const demos = world.meetings.filter(m => m.meeting_type === 'demo_day');
-  const oneonones = world.meetings.filter(m => m.meeting_type === 'one_on_one');
-  const boards = world.meetings.filter(m => m.meeting_type === 'board_meeting');
-
-  const subset: EntityFacts[] = [
-    ...founders.slice(0, 40),
-    ...partners.slice(0, 20),
-    ...engineers.slice(0, 10),
-    ...advisors.slice(0, 10),
-    ...startups.slice(0, 60),
-    ...vcs.slice(0, 15),
-    ...acquirers.slice(0, 5),
-    ...demos.slice(0, 15),
-    ...oneonones.slice(0, 25),
-    ...boards.slice(0, 10),
-    ...world.concepts.slice(0, 30),
-  ].slice(0, max);
-
-  console.log(`Generating ${subset.length} rich pages via Opus.`);
-  console.log(`Hard stop at $${HARD_STOP_USD}.`);
-  console.log(`Cache dir: ${OUTPUT_DIR}\n`);
-
-  if (dryRun) {
-    console.log('DRY RUN: would generate', subset.length, 'pages');
-    console.log('Distribution:', {
-      people: subset.filter(e => e.type === 'person').length,
-      companies: subset.filter(e => e.type === 'company').length,
-      meetings: subset.filter(e => e.type === 'meeting').length,
-      concepts: subset.filter(e => e.type === 'concept').length,
-    });
-    return;
-  }
-
-  console.log(`Concurrency: ${concurrency}\n`);
-
-  // Already-cached count up front, so progress is honest.
-  const preCached = subset.filter(e => existsSync(join(OUTPUT_DIR, `${e.slug.replace('/', '__')}.json`))).length;
-  const toGenerate = subset.length - preCached;
-  console.log(`Already cached: ${preCached}. To generate: ${toGenerate}.\n`);
-
-  const queue = [...subset];
-  let cached = 0, generated = 0, failed = 0;
-  const startTime = Date.now();
-
-  function reportProgress(slug: string) {
-    const elapsedSec = (Date.now() - startTime) / 1000;
-    const rate = generated > 0 ? generated / elapsedSec : 0; // pages/sec
-    const remaining = toGenerate - generated;
-    const etaSec = rate > 0 ? remaining / rate : 0;
-    const etaMin = etaSec / 60;
-    const avgCostPer = generated > 0 ? ledger.costUsd / generated : 0;
-    const projectedTotal = avgCostPer * toGenerate;
-    console.log(`  [${elapsedSec.toFixed(0)}s] ${generated}/${toGenerate} (${cached} cached) — $${ledger.costUsd.toFixed(2)} spent — rate ${rate.toFixed(2)}/s — ETA ${etaMin.toFixed(1)}min — projected total $${projectedTotal.toFixed(2)} — last: ${slug}`);
-  }
-
-  async function worker() {
-    while (queue.length > 0) {
-      const e = queue.shift();
-      if (!e) break;
-      const r = await generateOne(e, world);
-      if (r.ok) {
-        if (r.cached) cached++;
-        else {
-          generated++;
-          // Report on every page when sequential; every 5 when concurrent.
-          if (concurrency === 1 || generated % 5 === 0) reportProgress(e.slug);
-        }
-      } else {
-        failed++;
-        console.error(`  FAIL ${e.slug}: ${r.error}`);
-        if (r.error.startsWith('HARD_STOP')) {
-          queue.length = 0; // drain
-          return;
-        }
-      }
-    }
-  }
-
-  await Promise.all(Array.from({ length: concurrency }, () => worker()));
-
-  console.log(`\nDone. ${generated} generated, ${cached} cached, ${failed} failed.`);
-  console.log(`Total cost: $${ledger.costUsd.toFixed(2)} (${ledger.calls} calls, ${ledger.inputTokens.toLocaleString()} in / ${ledger.outputTokens.toLocaleString()} out)`);
-  console.log(`Output dir: ${OUTPUT_DIR}/`);
-
-  // Persist ledger for reproducibility.
-  writeFileSync(join(OUTPUT_DIR, '_ledger.json'), JSON.stringify({
-    generated_at: new Date().toISOString(),
-    model: MODEL,
-    pricing: { input_per_m: PRICE_INPUT_PER_M, output_per_m: PRICE_OUTPUT_PER_M },
-    ...ledger,
-    files_total: readdirSync(OUTPUT_DIR).filter(f => f.endsWith('.json') && f !== '_ledger.json').length,
-  }, null, 2));
-}
-
-main().catch(e => { console.error(e); process.exit(1); });
diff --git a/eval/generators/world.ts b/eval/generators/world.ts
deleted file mode 100644
index 73bb23e61..000000000
--- a/eval/generators/world.ts
+++ /dev/null
@@ -1,373 +0,0 @@
-/**
- * World skeleton generator (procedural, no LLM).
- *
- * Produces a coherent fictional VC-portfolio-style ecosystem:
- *   200 people (founders, partners, engineers, advisors)
- *   150 companies (startups, VCs, acquirers)
- *   100 meetings (demo days, 1:1s, board meetings, batch reviews)
- *   50 concepts (themes, frameworks)
- *
- * Each entity has structured "facts" — the ground truth we'll later use to
- * measure whether prose-generated content preserves enough signal for
- * extraction and search to recover the facts.
- *
- * Deterministic given a seed (reproducibility matters for benchmarks).
- */
-
-export type EntityType = 'person' | 'company' | 'meeting' | 'concept';
-
-export interface PersonFacts {
-  type: 'person';
-  slug: string;
-  name: string;
-  role: 'founder' | 'partner' | 'engineer' | 'advisor';
-  /** For founders: the company they founded. For employees: where they work. For advisors: companies they advise. For partners: companies they invested in. */
-  primary_affiliation: string; // company slug
-  secondary_affiliations?: string[]; // additional company slugs (advisors/engineers can have multiple)
-  notable_traits: string[]; // 2-3 traits to give the LLM something to work with
-  background?: string; // one-line bio seed
-}
-
-export interface CompanyFacts {
-  type: 'company';
-  slug: string;
-  name: string;
-  category: 'startup' | 'vc' | 'acquirer' | 'mature';
-  industry: string;
-  founded_year?: number;
-  /** Computed from people facts when consolidated. */
-  founders?: string[];
-  employees?: string[];
-  investors?: string[];
-  advisors?: string[];
-}
-
-export interface MeetingFacts {
-  type: 'meeting';
-  slug: string;
-  name: string;
-  meeting_type: 'demo_day' | 'one_on_one' | 'board_meeting' | 'batch_review';
-  date: string;
-  attendees: string[]; // people slugs
-  topic_company?: string; // company slug discussed
-  topic_concept?: string; // concept slug
-}
-
-export interface ConceptFacts {
-  type: 'concept';
-  slug: string;
-  name: string;
-  description: string;
-  related_companies: string[]; // company slugs that exemplify this concept
-  related_people?: string[]; // people most associated with this concept
-}
-
-export type EntityFacts = PersonFacts | CompanyFacts | MeetingFacts | ConceptFacts;
-
-// ─── Name pools ────────────────────────────────────────────────
-
-const FIRST_NAMES = [
-  'Sarah', 'Alice', 'Bob', 'Carol', 'David', 'Eve', 'Frank', 'Grace', 'Henry', 'Iris',
-  'Jack', 'Kate', 'Liam', 'Mia', 'Noah', 'Olivia', 'Paul', 'Quinn', 'Rachel', 'Sam',
-  'Tara', 'Uma', 'Victor', 'Wendy', 'Xavier', 'Yara', 'Zoe', 'Adam', 'Beth', 'Chris',
-  'Diana', 'Eric', 'Fiona', 'Gabe', 'Helen', 'Ian', 'Julia', 'Kevin', 'Linda', 'Mark',
-  'Nina', 'Owen', 'Priya', 'Quinten', 'Rosa', 'Steve', 'Tina', 'Ulrich', 'Vera', 'Will',
-];
-const LAST_NAMES = [
-  'Chen', 'Smith', 'Johnson', 'Williams', 'Brown', 'Jones', 'Garcia', 'Miller', 'Davis', 'Rodriguez',
-  'Martinez', 'Hernandez', 'Lopez', 'Gonzalez', 'Wilson', 'Anderson', 'Thomas', 'Taylor', 'Moore', 'Jackson',
-  'Lee', 'Park', 'Kim', 'Patel', 'Kapoor', 'Nakamura', 'Liu', 'Zhang', 'Wang', 'Singh',
-];
-const COMPANY_NAMES_STARTUP = [
-  'Acme', 'Beta', 'Gamma', 'Delta', 'Epsilon', 'Nimbus', 'Vector', 'Quantum', 'Pulse', 'Helix',
-  'Beacon', 'Compass', 'Lumen', 'Cipher', 'Mosaic', 'Tessera', 'Mantle', 'Gravity', 'Apex', 'Forge',
-  'Kindle', 'Lucid', 'Ranger', 'Sentinel', 'Tempo', 'Vox', 'Wisp', 'Zenith', 'Anchor', 'Brink',
-  'Cascade', 'Drift', 'Echo', 'Foundry', 'Gust', 'Hatch', 'Iris', 'Jolt', 'Keel', 'Lattice',
-  'Meridian', 'Nexus', 'Orbit', 'Prism', 'Quasar', 'Resonance', 'Spire', 'Talon', 'Umbra', 'Vellum',
-];
-const COMPANY_NAMES_VC = [
-  'Founders Fund', 'Sequoia Capital', 'Andreessen Horowitz', 'Benchmark', 'Greylock',
-  'Accel', 'Lightspeed', 'Index Ventures', 'Khosla Ventures', 'Floodgate',
-  'First Round', 'Initialized', 'Bessemer', 'NEA', 'Kleiner Perkins',
-];
-const COMPANY_NAMES_ACQUIRER = [
-  'Microsoft', 'Google', 'Meta', 'Amazon', 'Apple', 'Salesforce', 'Oracle', 'Adobe', 'Cisco', 'Intel',
-];
-const INDUSTRIES = [
-  'AI infrastructure', 'AI applications', 'fintech', 'climate tech', 'biotech',
-  'developer tools', 'enterprise SaaS', 'consumer social', 'crypto', 'robotics',
-  'edtech', 'health tech', 'cybersecurity', 'logistics', 'data infrastructure',
-];
-const PERSON_TRAITS = [
-  'product-obsessed', 'technical depth', 'fundraising-savvy', 'recruiting strength',
-  'design taste', 'analytical', 'opinionated', 'collaborative', 'fast-shipping',
-  'long-term thinker', 'demanding', 'patient', 'sharp pattern matcher', 'storyteller',
-  'first-principles thinker', 'systems builder', 'distribution-focused', 'GTM-heavy',
-];
-const CONCEPT_NAMES = [
-  'product-market fit', 'do things that don\'t scale', 'founder mode', 'second-time founder',
-  'AI-first product', 'unit economics', 'open source distribution', 'community-led growth',
-  'usage-based pricing', 'agentic workflows', 'foundation models', 'fine-tuning',
-  'retrieval augmented generation', 'multi-modal', 'inference cost', 'latency budget',
-  'customer concentration', 'revenue durability', 'gross margin expansion', 'churn cohorts',
-  'embedded fintech', 'wallet share', 'carbon credits', 'permitting reform',
-  'vertical SaaS', 'horizontal API', 'developer relations', 'PLG motion',
-  'enterprise GTM', 'top-down sales', 'bottom-up adoption', 'land and expand',
-  'category creation', 'platform shift', 'incumbent disruption', 'distribution moat',
-  'data moat', 'network effects', 'switching costs', 'pricing power',
-  'series A graduation', 'down round dynamics', 'secondary markets', 'liquidity events',
-  'M&A integration', 'cultural fit', 'remote-first', 'in-person culture',
-  'AI safety', 'alignment research', 'inference economics', 'training compute',
-];
-
-// ─── Deterministic RNG ────────────────────────────────────────
-
-function mulberry32(seed: number): () => number {
-  let t = seed >>> 0;
-  return () => {
-    t += 0x6D2B79F5;
-    let r = t;
-    r = Math.imul(r ^ (r >>> 15), r | 1);
-    r ^= r + Math.imul(r ^ (r >>> 7), r | 61);
-    return ((r ^ (r >>> 14)) >>> 0) / 4294967296;
-  };
-}
-
-function pick<T>(arr: T[], rand: () => number): T {
-  return arr[Math.floor(rand() * arr.length)];
-}
-
-function pickN<T>(arr: T[], n: number, rand: () => number): T[] {
-  const copy = [...arr];
-  const out: T[] = [];
-  for (let i = 0; i < n && copy.length > 0; i++) {
-    out.push(copy.splice(Math.floor(rand() * copy.length), 1)[0]);
-  }
-  return out;
-}
-
-function slugify(s: string): string {
-  return s.toLowerCase().replace(/[^a-z0-9]+/g, '-').replace(/^-|-$/g, '').slice(0, 60);
-}
-
-// ─── World construction ──────────────────────────────────────
-
-export interface World {
-  people: PersonFacts[];
-  companies: CompanyFacts[];
-  meetings: MeetingFacts[];
-  concepts: ConceptFacts[];
-}
-
-export function buildWorld(seed: number = 42): World {
-  const rand = mulberry32(seed);
-
-  // 1. Companies first (people reference them).
-  const companies: CompanyFacts[] = [];
-  // Startups: 100 (most numerous)
-  for (let i = 0; i < 100 && i < COMPANY_NAMES_STARTUP.length * 2; i++) {
-    const name = COMPANY_NAMES_STARTUP[i % COMPANY_NAMES_STARTUP.length] + (i >= COMPANY_NAMES_STARTUP.length ? ' Labs' : '');
-    companies.push({
-      type: 'company',
-      slug: `companies/${slugify(name)}-${i}`,
-      name,
-      category: 'startup',
-      industry: pick(INDUSTRIES, rand),
-      founded_year: 2018 + Math.floor(rand() * 8), // 2018-2025
-    });
-  }
-  // VCs: 25
-  for (let i = 0; i < 25; i++) {
-    const name = COMPANY_NAMES_VC[i % COMPANY_NAMES_VC.length] + (i >= COMPANY_NAMES_VC.length ? ' II' : '');
-    companies.push({
-      type: 'company',
-      slug: `companies/${slugify(name)}-${i}`,
-      name,
-      category: 'vc',
-      industry: 'venture capital',
-    });
-  }
-  // Acquirers + mature: 25
-  for (let i = 0; i < 25; i++) {
-    const name = i < COMPANY_NAMES_ACQUIRER.length ? COMPANY_NAMES_ACQUIRER[i] : COMPANY_NAMES_STARTUP[i] + ' Corp';
-    companies.push({
-      type: 'company',
-      slug: `companies/${slugify(name)}-${i}`,
-      name,
-      category: i < COMPANY_NAMES_ACQUIRER.length ? 'acquirer' : 'mature',
-      industry: pick(INDUSTRIES, rand),
-      founded_year: i < COMPANY_NAMES_ACQUIRER.length ? 1995 + i : 2010 + i,
-    });
-  }
-
-  const startupSlugs = companies.filter(c => c.category === 'startup').map(c => c.slug);
-  const vcSlugs = companies.filter(c => c.category === 'vc').map(c => c.slug);
-
-  // 2. People — wired to companies.
-  const people: PersonFacts[] = [];
-  let usedNames = new Set<string>();
-  function newName(): string {
-    while (true) {
-      const n = `${pick(FIRST_NAMES, rand)} ${pick(LAST_NAMES, rand)}`;
-      if (!usedNames.has(n)) { usedNames.add(n); return n; }
-    }
-  }
-
-  // 80 founders — one per startup that has founders.
-  for (let i = 0; i < 80; i++) {
-    const name = newName();
-    const company = startupSlugs[i % startupSlugs.length];
-    people.push({
-      type: 'person',
-      slug: `people/${slugify(name)}-${i}`,
-      name,
-      role: 'founder',
-      primary_affiliation: company,
-      notable_traits: pickN(PERSON_TRAITS, 2, rand),
-    });
-    const cf = companies.find(c => c.slug === company);
-    if (cf) (cf.founders ??= []).push(`people/${slugify(name)}-${i}`);
-  }
-  // 30 partners — at VCs.
-  for (let i = 0; i < 30; i++) {
-    const name = newName();
-    const vc = vcSlugs[i % vcSlugs.length];
-    // Each partner invests in 3-5 startups.
-    const investments = pickN(startupSlugs, 3 + Math.floor(rand() * 3), rand);
-    people.push({
-      type: 'person',
-      slug: `people/${slugify(name)}-${i + 80}`,
-      name,
-      role: 'partner',
-      primary_affiliation: vc,
-      secondary_affiliations: investments,
-      notable_traits: pickN(PERSON_TRAITS, 2, rand),
-    });
-    for (const s of investments) {
-      const cf = companies.find(c => c.slug === s);
-      if (cf) (cf.investors ??= []).push(`people/${slugify(name)}-${i + 80}`);
-    }
-  }
-  // 60 engineers — at startups.
-  for (let i = 0; i < 60; i++) {
-    const name = newName();
-    const employer = startupSlugs[i % startupSlugs.length];
-    const previous = startupSlugs[(i + 17) % startupSlugs.length]; // some have a prior gig
-    people.push({
-      type: 'person',
-      slug: `people/${slugify(name)}-${i + 110}`,
-      name,
-      role: 'engineer',
-      primary_affiliation: employer,
-      secondary_affiliations: rand() < 0.4 ? [previous] : [],
-      notable_traits: pickN(PERSON_TRAITS, 2, rand),
-    });
-    const cf = companies.find(c => c.slug === employer);
-    if (cf) (cf.employees ??= []).push(`people/${slugify(name)}-${i + 110}`);
-  }
-  // 30 advisors — cross-company.
-  for (let i = 0; i < 30; i++) {
-    const name = newName();
-    const advised = pickN(startupSlugs, 2 + Math.floor(rand() * 3), rand);
-    people.push({
-      type: 'person',
-      slug: `people/${slugify(name)}-${i + 170}`,
-      name,
-      role: 'advisor',
-      primary_affiliation: advised[0],
-      secondary_affiliations: advised.slice(1),
-      notable_traits: pickN(PERSON_TRAITS, 2, rand),
-    });
-    for (const s of advised) {
-      const cf = companies.find(c => c.slug === s);
-      if (cf) (cf.advisors ??= []).push(`people/${slugify(name)}-${i + 170}`);
-    }
-  }
-
-  // 3. Meetings — wire to people + companies.
-  const meetings: MeetingFacts[] = [];
-  const founders = people.filter(p => p.role === 'founder');
-  const partners = people.filter(p => p.role === 'partner');
-  const advisors = people.filter(p => p.role === 'advisor');
-  const engineers = people.filter(p => p.role === 'engineer');
-
-  // 30 demo days
-  for (let i = 0; i < 30; i++) {
-    const attendees = [
-      partners[i % partners.length].slug,
-      founders[i % founders.length].slug,
-      founders[(i + 5) % founders.length].slug,
-      founders[(i + 11) % founders.length].slug,
-      engineers[i % engineers.length].slug,
-    ];
-    meetings.push({
-      type: 'meeting',
-      slug: `meetings/demo-day-${2024 + Math.floor(i / 12)}-${String((i % 12) + 1).padStart(2, '0')}-${String(15 + i % 10).padStart(2, '0')}-batch-${i}`,
-      name: `Demo Day W${24 + i}`,
-      meeting_type: 'demo_day',
-      date: `${2024 + Math.floor(i / 12)}-${String((i % 12) + 1).padStart(2, '0')}-${String(15 + i % 10).padStart(2, '0')}`,
-      attendees,
-      topic_company: founders[i % founders.length].primary_affiliation,
-    });
-  }
-  // 40 1:1s
-  for (let i = 0; i < 40; i++) {
-    meetings.push({
-      type: 'meeting',
-      slug: `meetings/oneonone-${i}-${2025}-${String((i % 12) + 1).padStart(2, '0')}-${String((i % 28) + 1).padStart(2, '0')}`,
-      name: `1:1 ${partners[i % partners.length].name} + ${founders[i % founders.length].name}`,
-      meeting_type: 'one_on_one',
-      date: `2025-${String((i % 12) + 1).padStart(2, '0')}-${String((i % 28) + 1).padStart(2, '0')}`,
-      attendees: [partners[i % partners.length].slug, founders[i % founders.length].slug],
-      topic_company: founders[i % founders.length].primary_affiliation,
-    });
-  }
-  // 30 board meetings
-  for (let i = 0; i < 30; i++) {
-    const company = startupSlugs[i % startupSlugs.length];
-    const cf = companies.find(c => c.slug === company);
-    const attendees = [
-      ...(cf?.founders?.slice(0, 1) ?? []),
-      ...(cf?.investors?.slice(0, 2) ?? []),
-      ...(cf?.advisors?.slice(0, 1) ?? []),
-    ];
-    meetings.push({
-      type: 'meeting',
-      slug: `meetings/board-${slugify(cf?.name ?? 'unknown')}-${2025 + Math.floor(i / 12)}-q${(i % 4) + 1}-${i}`,
-      name: `${cf?.name} Board Meeting Q${(i % 4) + 1}`,
-      meeting_type: 'board_meeting',
-      date: `${2025 + Math.floor(i / 12)}-${String(((i % 4) * 3) + 1).padStart(2, '0')}-15`,
-      attendees,
-      topic_company: company,
-    });
-  }
-
-  // 4. Concepts.
-  const concepts: ConceptFacts[] = [];
-  for (let i = 0; i < Math.min(50, CONCEPT_NAMES.length); i++) {
-    const c = CONCEPT_NAMES[i];
-    concepts.push({
-      type: 'concept',
-      slug: `concepts/${slugify(c)}`,
-      name: c,
-      description: `${c} as a strategic frame for thinking about company building.`,
-      related_companies: pickN(startupSlugs, 3, rand),
-      related_people: pickN(people.map(p => p.slug), 2, rand),
-    });
-  }
-
-  return { people, companies, meetings, concepts };
-}
-
-// ─── Export to JSON for the gen pass ───
-
-if (import.meta.main) {
-  const world = buildWorld(42);
-  console.log(`World built: ${world.people.length} people, ${world.companies.length} companies, ${world.meetings.length} meetings, ${world.concepts.length} concepts`);
-  console.log(`Total entities: ${world.people.length + world.companies.length + world.meetings.length + world.concepts.length}`);
-  // Sample: show first of each type
-  console.log('\nSample person:', JSON.stringify(world.people[0], null, 2));
-  console.log('\nSample company:', JSON.stringify(world.companies[0], null, 2));
-  console.log('\nSample meeting:', JSON.stringify(world.meetings[0], null, 2));
-  console.log('\nSample concept:', JSON.stringify(world.concepts[0], null, 2));
-}
diff --git a/eval/runner/adversarial.ts b/eval/runner/adversarial.ts
deleted file mode 100644
index 88b8cf125..000000000
--- a/eval/runner/adversarial.ts
+++ /dev/null
@@ -1,262 +0,0 @@
-/**
- * BrainBench Category 10: Robustness / Adversarial.
- *
- * Tests that gbrain doesn't crash, hang, or silently corrupt on weird input.
- * Currently zero coverage of edge cases. Pass = no exceptions, no hangs > 30s,
- * no silent wrong-data outputs.
- *
- * Usage: bun run eval/runner/adversarial.ts [--json]
- */
-
-import { PGLiteEngine } from '../../src/core/pglite-engine.ts';
-import { extractPageLinks, parseTimelineEntries } from '../../src/core/link-extraction.ts';
-import type { PageInput } from '../../src/core/types.ts';
-
-interface CaseResult {
-  name: string;
-  ops_attempted: number;
-  ops_succeeded: number;
-  crashes: string[];
-  silent_corruption: string[];
-}
-
-interface AdversarialCase {
-  name: string;
-  slug: string;
-  page: PageInput;
-  /** Optional invariants to check after putPage. */
-  expect?: (engine: PGLiteEngine) => Promise<{ pass: boolean; note?: string }>;
-}
-
-const ADVERSARIAL_CASES: AdversarialCase[] = [
-  // ── Empty / whitespace ──
-  { name: 'empty compiled_truth', slug: 'concepts/empty-1', page: { type: 'concept', title: '', compiled_truth: '', timeline: '' } },
-  { name: 'whitespace only', slug: 'concepts/ws-1', page: { type: 'concept', title: '   ', compiled_truth: '\n\t  \n', timeline: '   ' } },
-  { name: 'newlines only', slug: 'concepts/nl-1', page: { type: 'concept', title: 'NL', compiled_truth: '\n\n\n\n', timeline: '\n\n' } },
-
-  // ── Massive content ──
-  { name: '50K char page', slug: 'concepts/big-1', page: { type: 'concept', title: 'Big', compiled_truth: 'Lorem ipsum dolor sit amet, '.repeat(2000), timeline: '' } },
-  { name: '100K char page', slug: 'concepts/huge-1', page: { type: 'concept', title: 'Huge', compiled_truth: 'A'.repeat(100_000), timeline: '' } },
-
-  // ── Unicode / non-Latin ──
-  { name: 'CJK content', slug: 'concepts/cjk-1', page: { type: 'concept', title: '人工智能', compiled_truth: '这是一个关于人工智能的页面。提到了 [公司](companies/acme)。', timeline: '- **2026-01-01** | 创立' } },
-  { name: 'Arabic RTL', slug: 'concepts/ar-1', page: { type: 'concept', title: 'الذكاء', compiled_truth: 'هذه صفحة عن الذكاء الاصطناعي.', timeline: '' } },
-  { name: 'Cyrillic', slug: 'concepts/cy-1', page: { type: 'concept', title: 'Привет', compiled_truth: 'Это страница о технологиях.', timeline: '' } },
-  { name: 'emoji-heavy', slug: 'concepts/emoji-1', page: { type: 'concept', title: '🚀 Launch', compiled_truth: '🚀🎉🔥 Launched! 💯 [Acme](companies/acme) 👏👏👏', timeline: '- **2026-01-01** | 🚀 launched' } },
-  { name: 'mixed scripts', slug: 'concepts/mixed-1', page: { type: 'concept', title: 'Mix 混合 العربية', compiled_truth: 'English 中文 العربية русский 🇺🇸 [Acme](companies/acme).', timeline: '' } },
-
-  // ── Code fences (slugs inside MUST NOT extract) ──
-  { name: 'slug inside code fence', slug: 'concepts/code-1', page: {
-    type: 'concept', title: 'Code',
-    compiled_truth: 'See this code:\n```\nconst x = "people/should-not-extract";\nconst y = [Test](people/also-not-extract);\n```\nReal ref: [Real](people/real-target).',
-    timeline: '',
-  } },
-  { name: 'inline code with slug', slug: 'concepts/inline-1', page: {
-    type: 'concept', title: 'Inline',
-    compiled_truth: 'Use the `people/code-fenced-slug` notation. Real ref: [Real](people/real-target-2).',
-    timeline: '',
-  } },
-
-  // ── False-positive substrings ──
-  { name: 'false-positive substring', slug: 'concepts/fp-1', page: {
-    type: 'concept', title: 'FP',
-    compiled_truth: 'A frank discussion of founder mode is needed. Note that [frank-founder](people/frank-founder) attended.',
-    timeline: '',
-  } },
-
-  // ── Slugs with edge characters ──
-  { name: 'slug with dots', slug: 'concepts/dot.in.slug', page: { type: 'concept', title: 'Dotty', compiled_truth: 'A page.', timeline: '' } },
-  { name: 'slug with leading number', slug: 'concepts/123-numeric', page: { type: 'concept', title: 'Numeric', compiled_truth: 'A page.', timeline: '' } },
-  { name: 'slug max length', slug: 'concepts/' + 'x'.repeat(200), page: { type: 'concept', title: 'Long', compiled_truth: 'A page.', timeline: '' } },
-
-  // ── Malformed timeline ──
-  { name: 'invalid date in timeline', slug: 'concepts/bad-date-1', page: {
-    type: 'concept', title: 'BadDate',
-    compiled_truth: 'A page.',
-    timeline: '- **2026-13-45** | Invalid date\n- **not-a-date** | Garbage\n- **2026-02-15** | Valid entry',
-  } },
-  { name: 'timeline with no dates', slug: 'concepts/no-dates-1', page: {
-    type: 'concept', title: 'NoDates',
-    compiled_truth: 'A page.',
-    timeline: '- Just a bullet, no date\n- Another bullet',
-  } },
-
-  // ── Deeply nested markdown ──
-  { name: 'deeply nested lists', slug: 'concepts/nested-1', page: {
-    type: 'concept', title: 'Nested',
-    compiled_truth: '- L1\n  - L2\n    - L3\n      - L4\n        - L5\n          - L6\n            - L7\n              - L8',
-    timeline: '',
-  } },
-  { name: 'long blockquote chain', slug: 'concepts/quote-1', page: {
-    type: 'concept', title: 'Quoted',
-    compiled_truth: '> > > > > > deeply quoted [Acme](companies/acme).',
-    timeline: '',
-  } },
-
-  // ── Many entity refs in one page ──
-  { name: '100 refs in one page', slug: 'meetings/megamention-1', page: {
-    type: 'meeting', title: 'Mega',
-    compiled_truth: Array.from({ length: 100 }, (_, i) => `[Person ${i}](people/p-${i})`).join(', '),
-    timeline: '',
-  } },
-
-  // ── Repeated mentions of same entity (should dedupe to 1 link) ──
-  // Note: tests extractPageLinks directly. engine.putPage doesn't run auto-link;
-  // the operation handler does. Within-page dedup happens inside extractPageLinks.
-  { name: 'same entity 50 times', slug: 'concepts/repeat-1', page: {
-    type: 'concept', title: 'Repeat',
-    compiled_truth: Array.from({ length: 50 }, () => '[Same](people/same-target)').join(' '),
-    timeline: '',
-  }, expect: async () => {
-    const candidates = extractPageLinks(
-      Array.from({ length: 50 }, () => '[Same](people/same-target)').join(' '),
-      {},
-      'concept',
-    );
-    const matches = candidates.filter(c => c.targetSlug === 'people/same-target');
-    return { pass: matches.length === 1, note: `expected 1 candidate after within-page dedup, got ${matches.length}` };
-  } },
-];
-
-async function tryOp<T>(name: string, fn: () => Promise<T>): Promise<{ ok: true; result: T } | { ok: false; error: string }> {
-  try {
-    const r = await Promise.race([
-      fn().then(x => ({ ok: true as const, value: x })),
-      new Promise<{ ok: false; error: string }>((_, reject) =>
-        setTimeout(() => reject({ ok: false, error: 'TIMEOUT_30s' }), 30_000),
-      ),
-    ]);
-    if ('value' in r && r.ok) return { ok: true, result: r.value };
-    return { ok: false, error: 'unknown' };
-  } catch (e) {
-    const err = e instanceof Error ? e.message : String(e);
-    return { ok: false, error: `${name}: ${err.slice(0, 200)}` };
-  }
-}
-
-async function main() {
-  const json = process.argv.includes('--json');
-  const log = json ? () => {} : console.log;
-
-  log('# BrainBench Category 10: Robustness / Adversarial\n');
-  log(`Generated: ${new Date().toISOString().slice(0, 19)}`);
-
-  const engine = new PGLiteEngine();
-  await engine.connect({});
-  await engine.initSchema();
-
-  const targetPages = [
-    { slug: 'people/real-target', page: { type: 'person' as const, title: 'Real', compiled_truth: 'A real person.', timeline: '' } },
-    { slug: 'people/real-target-2', page: { type: 'person' as const, title: 'Real2', compiled_truth: 'Real2.', timeline: '' } },
-    { slug: 'people/frank-founder', page: { type: 'person' as const, title: 'Frank', compiled_truth: 'A founder.', timeline: '' } },
-    { slug: 'people/same-target', page: { type: 'person' as const, title: 'Same', compiled_truth: 'Same.', timeline: '' } },
-    { slug: 'companies/acme', page: { type: 'company' as const, title: 'Acme', compiled_truth: 'Acme.', timeline: '' } },
-  ];
-  for (const tp of targetPages) await engine.putPage(tp.slug, tp.page);
-
-  const results: CaseResult[] = [];
-
-  for (const c of ADVERSARIAL_CASES) {
-    log(`\n## Case: ${c.name}`);
-    const result: CaseResult = { name: c.name, ops_attempted: 0, ops_succeeded: 0, crashes: [], silent_corruption: [] };
-
-    // 1. putPage
-    result.ops_attempted++;
-    const put = await tryOp('putPage', () => engine.putPage(c.slug, c.page));
-    if (put.ok) result.ops_succeeded++; else result.crashes.push(put.error);
-
-    // 2. getPage roundtrip — content should match what we put
-    result.ops_attempted++;
-    const got = await tryOp('getPage', () => engine.getPage(c.slug));
-    if (got.ok) {
-      result.ops_succeeded++;
-      const page = got.result;
-      if (page && page.compiled_truth !== c.page.compiled_truth) {
-        result.silent_corruption.push(`getPage roundtrip differs (${(page.compiled_truth ?? '').length} vs ${c.page.compiled_truth.length} chars)`);
-      }
-    } else result.crashes.push(got.error);
-
-    // 3. searchKeyword
-    result.ops_attempted++;
-    const search = await tryOp('searchKeyword', () => engine.searchKeyword('person', { limit: 10 }));
-    if (search.ok) result.ops_succeeded++; else result.crashes.push(search.error);
-
-    // 4. extractPageLinks (pure function, code-fence and false-positive checks happen here)
-    result.ops_attempted++;
-    const extract = await tryOp('extractPageLinks', async () => extractPageLinks(c.page.compiled_truth, {}, c.page.type));
-    if (extract.ok) {
-      result.ops_succeeded++;
-      // Check: code fence content should NOT produce link candidates
-      if (c.name.includes('code fence') || c.name.includes('inline code')) {
-        const candidates = extract.result;
-        const leaked = candidates.filter(cand => cand.targetSlug.includes('not-extract') || cand.targetSlug.includes('code-fenced-slug'));
-        if (leaked.length > 0) result.silent_corruption.push(`code fence leak: extracted ${leaked.map(l => l.targetSlug).join(', ')}`);
-      }
-    } else result.crashes.push(extract.error);
-
-    // 5. parseTimelineEntries
-    result.ops_attempted++;
-    const tl = await tryOp('parseTimelineEntries', async () => parseTimelineEntries(c.page.timeline ?? ''));
-    if (tl.ok) {
-      result.ops_succeeded++;
-      // For "invalid date" case, valid entries should still parse
-      if (c.name === 'invalid date in timeline') {
-        const valid = tl.result.filter(e => e.date === '2026-02-15');
-        if (valid.length === 0) result.silent_corruption.push('valid entry lost when other entries had invalid dates');
-      }
-    } else result.crashes.push(tl.error);
-
-    // 6. traversePaths from this slug
-    result.ops_attempted++;
-    const traverse = await tryOp('traversePaths', () => engine.traversePaths(c.slug, { depth: 2 }));
-    if (traverse.ok) result.ops_succeeded++; else result.crashes.push(traverse.error);
-
-    // 7. Custom expect
-    if (c.expect) {
-      result.ops_attempted++;
-      const expectResult = await tryOp('expect', () => c.expect!(engine));
-      if (expectResult.ok) {
-        result.ops_succeeded++;
-        if (!expectResult.result.pass) {
-          result.silent_corruption.push(`invariant failed: ${expectResult.result.note ?? 'no note'}`);
-        }
-      } else result.crashes.push(expectResult.error);
-    }
-
-    log(`  ${result.ops_succeeded}/${result.ops_attempted} ops succeeded`);
-    if (result.crashes.length > 0) log(`  ✗ crashes: ${result.crashes.length}`);
-    if (result.silent_corruption.length > 0) log(`  ✗ silent corruption: ${result.silent_corruption.length}`);
-    for (const c2 of result.crashes) log(`    crash: ${c2}`);
-    for (const sc of result.silent_corruption) log(`    silent: ${sc}`);
-
-    results.push(result);
-  }
-
-  await engine.disconnect();
-
-  const totalOps = results.reduce((s, r) => s + r.ops_attempted, 0);
-  const totalSucc = results.reduce((s, r) => s + r.ops_succeeded, 0);
-  const totalCrashes = results.reduce((s, r) => s + r.crashes.length, 0);
-  const totalSilent = results.reduce((s, r) => s + r.silent_corruption.length, 0);
-
-  log(`\n## Summary`);
-  log(`Cases: ${results.length}`);
-  log(`Ops attempted: ${totalOps}`);
-  log(`Ops succeeded: ${totalSucc} (${((totalSucc / totalOps) * 100).toFixed(1)}%)`);
-  log(`Crashes: ${totalCrashes}`);
-  log(`Silent corruption: ${totalSilent}`);
-
-  if (json) {
-    process.stdout.write(JSON.stringify({ results, summary: { totalOps, totalSucc, totalCrashes, totalSilent } }, null, 2) + '\n');
-  }
-
-  if (totalCrashes > 0 || totalSilent > 0) {
-    console.error(`\n⚠ ${totalCrashes} crash(es) and ${totalSilent} silent corruption(s) — see details above`);
-    process.exit(1);
-  }
-}
-
-main().catch(e => {
-  console.error('Adversarial eval error:', e);
-  process.exit(1);
-});
diff --git a/eval/runner/all.ts b/eval/runner/all.ts
deleted file mode 100644
index aeb054621..000000000
--- a/eval/runner/all.ts
+++ /dev/null
@@ -1,199 +0,0 @@
-/**
- * BrainBench v1 — combined runner.
- *
- * Runs every shipping eval category in sequence and writes a unified report
- * to eval/reports/YYYY-MM-DD-brainbench.md. Each category's full output is
- * captured and embedded in the report.
- *
- * Usage: bun run eval/runner/all.ts
- */
-
-import { execSync } from 'child_process';
-import { writeFileSync, mkdirSync, existsSync } from 'fs';
-import { join } from 'path';
-
-interface CategoryRun {
-  num: number;
-  name: string;
-  script: string;
-  status: 'pass' | 'fail';
-  output: string;
-  exitCode: number;
-}
-
-// One row per benchmark. Headline (Cat 1+2 combined) is the consolidated
-// before/after run on the full 240-page rich-prose corpus. Procedural
-// categories (3, 4, 7, 10, 12) test orthogonal capabilities.
-const CATEGORIES = [
-  { num: 1, name: 'Before/After PR #188 (240-page rich corpus, relational queries)', script: 'eval/runner/before-after.ts' },
-  { num: 3, name: 'Identity Resolution', script: 'eval/runner/identity.ts' },
-  { num: 4, name: 'Temporal Queries', script: 'eval/runner/temporal.ts' },
-  { num: 7, name: 'Performance / Latency', script: 'eval/runner/perf.ts' },
-  { num: 10, name: 'Robustness / Adversarial', script: 'eval/runner/adversarial.ts' },
-  { num: 12, name: 'MCP Operation Contract', script: 'eval/runner/mcp-contract.ts' },
-];
-
-function runCategory(c: typeof CATEGORIES[0]): CategoryRun {
-  console.log(`\n=== Running Category ${c.num}: ${c.name} ===`);
-  let output = '';
-  let exitCode = 0;
-  try {
-    output = execSync(`bun ${c.script}`, { encoding: 'utf-8', timeout: 600_000, maxBuffer: 50 * 1024 * 1024 });
-  } catch (e: unknown) {
-    const err = e as { stdout?: string; stderr?: string; status?: number };
-    output = (err.stdout || '') + (err.stderr || '');
-    exitCode = err.status ?? 1;
-  }
-  const lastLines = output.split('\n').slice(-5).join('\n');
-  console.log(lastLines);
-  return {
-    num: c.num,
-    name: c.name,
-    script: c.script,
-    status: exitCode === 0 ? 'pass' : 'fail',
-    output,
-    exitCode,
-  };
-}
-
-function buildReport(runs: CategoryRun[]): string {
-  const date = new Date().toISOString().slice(0, 10);
-  const passed = runs.filter(r => r.status === 'pass').length;
-  const failed = runs.length - passed;
-
-  const lines: string[] = [];
-  lines.push(`# BrainBench v1 — ${date}`);
-  lines.push('');
-  lines.push(`**Branch:** ${execSync('git rev-parse --abbrev-ref HEAD').toString().trim()}`);
-  lines.push(`**Commit:** \`${execSync('git rev-parse --short HEAD').toString().trim()}\``);
-  lines.push(`**Engine:** PGLite (in-memory)`);
-  lines.push('');
-
-  lines.push(`## Summary`);
-  lines.push('');
-  lines.push(`${runs.length} categories run. ${passed} passed, ${failed} failed.`);
-  lines.push('');
-  lines.push(`| # | Category | Status | Script |`);
-  lines.push(`|---|----------|--------|--------|`);
-  for (const r of runs) {
-    lines.push(`| ${r.num} | ${r.name} | ${r.status === 'pass' ? '✓ pass' : '✗ fail'} | \`${r.script}\` |`);
-  }
-  lines.push('');
-
-  lines.push(`## What this benchmark proves`);
-  lines.push('');
-  lines.push('BrainBench v1 evaluates gbrain end-to-end on a 240-page rich-prose corpus');
-  lines.push('(generated by Claude Opus 4.7, ~$15 one-time, committed to the repo).');
-  lines.push('The headline benchmark is a single before/after comparison: pre-PR-#188');
-  lines.push('vs the full v0.10.3 + v0.10.4 stack on the same data, same queries.');
-  lines.push('Reproducible: `bun run eval/runner/all.ts`, in-memory PGLite, no API keys');
-  lines.push('at run time, ~3 min total.');
-  lines.push('');
-  lines.push('### Headline: PR #188 strictly dominates baseline on every metric');
-  lines.push('');
-  lines.push('Real agents read ranked top-K results, not full sets. AFTER ranks graph');
-  lines.push('hits first (high precision), then fills with grep results. Both metrics');
-  lines.push('go UP — no category goes down.');
-  lines.push('');
-  lines.push('| Metric                | BEFORE PR #188 | AFTER PR #188 | Δ            |');
-  lines.push('|-----------------------|----------------|---------------|--------------|');
-  lines.push('| **Precision@5**       | **39.2%**      | **44.7%**     | **+5.4 pts** |');
-  lines.push('| **Recall@5**          | **83.1%**      | **94.6%**     | **+11.5 pts**|');
-  lines.push('| Correct in top-5      | 217            | 247           | **+30**      |');
-  lines.push('');
-  lines.push('Thirty more correct answers in the top-5 the agent actually reads. Recall');
-  lines.push('jumps 11.5 points — agents now find the answer in their first reads instead');
-  lines.push('of digging through grep noise.');
-  lines.push('');
-  lines.push('### Graph-only ablation (the typed graph alone, no grep)');
-  lines.push('');
-  lines.push('| Metric              | BEFORE (grep) | Graph-only  | Δ           |');
-  lines.push('|---------------------|---------------|-------------|-------------|');
-  lines.push('| **F1 score**        | 57.8%         | **86.6%**   | **+28.8 pts** |');
-  lines.push('| Set precision       | 40.8%         | **81.0%**   | **+40.2 pts** |');
-  lines.push('| Set recall          | 98.9%         | 93.1%       | -5.8 pts    |');
-  lines.push('| Total returned      | 632           | 300         | -53%        |');
-  lines.push('| Correct returned    | 258           | 243         | -6%         |');
-  lines.push('');
-  lines.push('Graph alone catches 94% of what grep catches with HALF the noise. The');
-  lines.push('5.8pt recall gap is mostly Opus-generated prose that paraphrases names');
-  lines.push('without markdown links ("Mark Thomas was there" instead of `[Mark Thomas](slug)`).');
-  lines.push('Closing this needs corpus-aware NER, deferred to v0.10.5.');
-  lines.push('');
-  lines.push('Per-link-type breakdown (graph-only):');
-  lines.push('');
-  lines.push('| Link type   | Expected | Graph found / returned | Recall  | Precision |');
-  lines.push('|-------------|----------|------------------------|---------|-----------|');
-  lines.push('| attended    | 134      | 131 / 134              | 97.8%   | 97.8%     |');
-  lines.push('| works_at    | 50       | 50 / 79                | 100.0%  | 63.3%     |');
-  lines.push('| invested_in | 60       | 50 / 56                | 83.3%   | 89.3%     |');
-  lines.push('| advises     | 17       | 12 / 31                | 70.6%   | 38.7%     |');
-  lines.push('');
-  lines.push('Graph wins biggest where grep is noisiest: relational questions on companies');
-  lines.push('with many incidental mentions ("Who works at Acme?" — grep returns every');
-  lines.push('page that mentions Acme; graph returns just the employees).');
-  lines.push('');
-  lines.push('## Categories not in v1 headline (deferred to v1.1, see TODOS.md)');
-  lines.push('- Cat 5: Source Attribution / Provenance');
-  lines.push('- Cat 6: Auto-link Precision under Prose (at scale beyond 240)');
-  lines.push('- Cat 8: Skill Behavior Compliance (needs LLM agent loop, ~$2K)');
-  lines.push('- Cat 9: End-to-End Workflows (needs LLM agent loop)');
-  lines.push('- Cat 11: Multi-modal Ingestion (needs licensed real datasets)');
-  lines.push('');
-
-  for (const r of runs) {
-    lines.push(`---`);
-    lines.push(`# Category ${r.num}: ${r.name}`);
-    lines.push('');
-    lines.push(`Status: ${r.status === 'pass' ? '✓ PASS' : '✗ FAIL'} (exit ${r.exitCode})`);
-    lines.push('');
-    lines.push('```');
-    // Trim setup noise (migration messages) from the head.
-    const trimmed = r.output
-      .split('\n')
-      .filter(l => !l.includes('Migration') || l.includes('Migration'))
-      .filter(l => !l.match(/^\s*\d+ migration\(s\) applied$/))
-      .join('\n');
-    lines.push(trimmed);
-    lines.push('```');
-    lines.push('');
-  }
-
-  lines.push(`---`);
-  lines.push(`## How to reproduce`);
-  lines.push('');
-  lines.push('```bash');
-  lines.push('bun run eval/runner/all.ts');
-  lines.push('```');
-  lines.push('');
-  lines.push('Each category can also run individually:');
-  lines.push('```bash');
-  for (const c of CATEGORIES) {
-    lines.push(`bun ${c.script}`);
-  }
-  lines.push('```');
-  lines.push('');
-  lines.push('No API keys required. All runs against PGLite in-memory. Total runtime ~3 min.');
-
-  return lines.join('\n');
-}
-
-async function main() {
-  const runs: CategoryRun[] = [];
-  for (const c of CATEGORIES) {
-    runs.push(runCategory(c));
-  }
-
-  const reportDir = 'eval/reports';
-  if (!existsSync(reportDir)) mkdirSync(reportDir, { recursive: true });
-  const date = new Date().toISOString().slice(0, 10);
-  const reportPath = join(reportDir, `${date}-brainbench.md`);
-  writeFileSync(reportPath, buildReport(runs));
-
-  console.log(`\n=== Report written to ${reportPath} ===`);
-  console.log(`${runs.filter(r => r.status === 'pass').length}/${runs.length} categories passed`);
-
-  if (runs.some(r => r.status === 'fail')) process.exit(1);
-}
-
-main().catch(e => { console.error(e); process.exit(1); });
diff --git a/eval/runner/before-after.ts b/eval/runner/before-after.ts
deleted file mode 100644
index cadb9768a..000000000
--- a/eval/runner/before-after.ts
+++ /dev/null
@@ -1,431 +0,0 @@
-/**
- * BrainBench v1 — single before/after comparison on the 240-page rich-prose corpus.
- *
- * Runs the same realistic synthetic brain through TWO configurations:
- *   BEFORE: gbrain pre-PR-#188. No auto-link, no extract --source db, no
- *           traversePaths, no backlink boost. Just put_page + searchKeyword
- *           and content-scan fallback for relational questions. This is what
- *           a vanilla v0.10.0 install does.
- *   AFTER:  gbrain after PR #188. Full graph layer: extract --source db
- *           populates typed links, traversePaths answers relational queries
- *           directly, backlink boost reranks search results, v0.10.4 prose
- *           regex fixes lift type accuracy from 70.7% → 88.5%.
- *
- * Same data. Same queries. Honest A/B numbers.
- *
- * Usage: bun eval/runner/before-after.ts [--json]
- */
-
-import { PGLiteEngine } from '../../src/core/pglite-engine.ts';
-import { runExtract } from '../../src/commands/extract.ts';
-import { readdirSync, readFileSync } from 'fs';
-import { join } from 'path';
-
-interface RichPage {
-  slug: string;
-  type: 'person' | 'company' | 'meeting' | 'concept';
-  title: string;
-  compiled_truth: string;
-  timeline: string;
-  _facts: {
-    type: string;
-    name?: string;
-    role?: string;
-    industry?: string;
-    primary_affiliation?: string;
-    secondary_affiliations?: string[];
-    founders?: string[];
-    employees?: string[];
-    investors?: string[];
-    advisors?: string[];
-    attendees?: string[];
-    related_companies?: string[];
-  };
-}
-
-function loadCorpus(dir: string): RichPage[] {
-  const files = readdirSync(dir).filter(f => f.endsWith('.json') && !f.startsWith('_'));
-  const out: RichPage[] = [];
-  for (const f of files) {
-    const p = JSON.parse(readFileSync(join(dir, f), 'utf-8'));
-    if (Array.isArray(p.timeline)) p.timeline = p.timeline.join('\n');
-    if (Array.isArray(p.compiled_truth)) p.compiled_truth = p.compiled_truth.join('\n\n');
-    p.title = String(p.title ?? '');
-    p.compiled_truth = String(p.compiled_truth ?? '');
-    p.timeline = String(p.timeline ?? '');
-    out.push(p as RichPage);
-  }
-  return out;
-}
-
-interface RelationalQuery {
-  question: string;
-  /** Source slug. */
-  seed: string;
-  /** Expected answer slugs. */
-  expected: string[];
-  /** Direction of relationship (in: who points at seed; out: what does seed point at). */
-  direction: 'in' | 'out';
-  /** Accept any of these link types as a match. ["works_at", "founded"]
-   *  for "who works at X" because founders are employees. */
-  linkTypes: string[];
-}
-
-function buildRelationalQueries(pages: RichPage[]): RelationalQuery[] {
-  const queries: RelationalQuery[] = [];
-  // Only entities that actually have generated pages are valid expected
-  // answers. The world generator references some entities (by slug in facts)
-  // that aren't in the 240-page Opus subset — those can't be extracted as
-  // links because the FK constraint blocks unresolved targets.
-  const existingSlugs = new Set(pages.map(p => p.slug));
-  const filterExisting = (slugs: string[]) => slugs.filter(s => existingSlugs.has(s));
-
-  // "Who attended meeting X?" — outgoing from each meeting page.
-  for (const p of pages) {
-    if (p._facts.type === 'meeting' && p._facts.attendees && p._facts.attendees.length > 0) {
-      const expected = filterExisting(p._facts.attendees);
-      if (expected.length === 0) continue;
-      queries.push({
-        question: `Who attended ${p.title}?`,
-        seed: p.slug,
-        expected,
-        direction: 'out',
-        linkTypes: ['attended'],
-      });
-    }
-  }
-
-  // "Who works at company X?" — incoming to each company.
-  // Founders are employees too — accept both `works_at` and `founded`.
-  for (const p of pages) {
-    if (p._facts.type === 'company' && p._facts.employees && p._facts.employees.length > 0) {
-      const expected = filterExisting([...(p._facts.employees ?? []), ...(p._facts.founders ?? [])]);
-      if (expected.length === 0) continue;
-      queries.push({
-        question: `Who works at ${p.title}?`,
-        seed: p.slug,
-        expected: [...new Set(expected)],
-        direction: 'in',
-        linkTypes: ['works_at', 'founded'],
-      });
-    }
-  }
-
-  // "Who invested in company X?" — incoming.
-  for (const p of pages) {
-    if (p._facts.type === 'company' && p._facts.investors && p._facts.investors.length > 0) {
-      const expected = filterExisting(p._facts.investors);
-      if (expected.length === 0) continue;
-      queries.push({
-        question: `Who invested in ${p.title}?`,
-        seed: p.slug,
-        expected,
-        direction: 'in',
-        linkTypes: ['invested_in'],
-      });
-    }
-  }
-
-  // "Who advises company X?"
-  for (const p of pages) {
-    if (p._facts.type === 'company' && p._facts.advisors && p._facts.advisors.length > 0) {
-      const expected = filterExisting(p._facts.advisors);
-      if (expected.length === 0) continue;
-      queries.push({
-        question: `Who advises ${p.title}?`,
-        seed: p.slug,
-        expected,
-        direction: 'in',
-        linkTypes: ['advises'],
-      });
-    }
-  }
-
-  return queries;
-}
-
-interface QueryResult {
-  question: string;
-  expected: number;
-  beforeFound: number;
-  beforeReturned: number;
-  /** Graph-only: typed traversal alone (precise but extraction-bound recall). */
-  graphOnlyFound: number;
-  graphOnlyReturned: number;
-  /** Hybrid: graph first, grep fallback for entities graph missed. */
-  afterFound: number;
-  afterReturned: number;
-  /** Top-K metrics — what the agent actually reads. */
-  beforeFoundAtK: number;   // correct in top-K BEFORE
-  afterFoundAtK: number;    // correct in top-K AFTER (graph-first ranking)
-}
-
-const TOP_K = 5;
-
-const ENTITY_REF_RE = /\[[^\]]+\]\(([^)]+)\)|\b((?:people|companies|meetings|concepts)\/[a-z0-9-]+)\b/gi;
-
-/** Pre-PR-188 fallback: extract entity refs from the seed page (outgoing) or
- *  scan all pages for the seed slug (incoming). This is what an agent on
- *  v0.10.0 would do — no graph, just text. */
-function beforePrAnswer(q: RelationalQuery, contentBySlug: Map<string, string>): Set<string> {
-  const returned = new Set<string>();
-  if (q.direction === 'out') {
-    const content = contentBySlug.get(q.seed) ?? '';
-    for (const m of content.matchAll(ENTITY_REF_RE)) {
-      const ref = (m[1] ?? m[2] ?? '').replace(/\.md$/, '').replace(/^\.\.\//, '');
-      if (ref && ref.includes('/') && ref !== q.seed) returned.add(ref);
-    }
-  } else {
-    // Incoming: grep all pages for seed slug.
-    for (const [slug, content] of contentBySlug) {
-      if (slug === q.seed) continue;
-      if (content.includes(q.seed)) returned.add(slug);
-    }
-  }
-  return returned;
-}
-
-async function main() {
-  const json = process.argv.includes('--json');
-  const log = json ? () => {} : console.log;
-
-  log('# BrainBench v1 — before/after PR #188\n');
-  log(`Generated: ${new Date().toISOString().slice(0, 19)}`);
-
-  const dir = 'eval/data/world-v1';
-  const pages = loadCorpus(dir);
-  log(`Corpus: ${pages.length} rich-prose pages from ${dir}/`);
-
-  const queries = buildRelationalQueries(pages);
-  log(`Relational queries: ${queries.length}`);
-
-  // ── BEFORE: just text ──
-  const contentBySlug = new Map<string, string>();
-  for (const p of pages) {
-    contentBySlug.set(p.slug, `${p.title}\n${p.compiled_truth}\n${p.timeline}`);
-  }
-
-  // ── AFTER: full graph layer ──
-  const engine = new PGLiteEngine();
-  await engine.connect({});
-  await engine.initSchema();
-
-  log('\n## Seeding corpus + running extract (v0.10.4 stack)');
-  for (const p of pages) {
-    await engine.putPage(p.slug, {
-      type: p.type,
-      title: p.title,
-      compiled_truth: p.compiled_truth,
-      timeline: p.timeline,
-    });
-  }
-  const captureLog = console.error;
-  console.error = () => {};
-  try {
-    await runExtract(engine, ['links', '--source', 'db']);
-    await runExtract(engine, ['timeline', '--source', 'db']);
-  } finally {
-    console.error = captureLog;
-  }
-  const stats = await engine.getStats();
-  log(`After extract: ${stats.link_count} typed links, ${stats.timeline_entry_count} timeline entries`);
-
-  // ── Run all queries through both configs ──
-  // BEFORE = grep-only fallback (what a v0.10.0 agent does)
-  // AFTER  = graph traversal + grep fallback for entities graph missed.
-  //   This is the realistic post-PR-#188 agent: it has BOTH tools and uses
-  //   them together. Graph results come first (high precision), grep fills
-  //   in entities the extractor missed (preserves recall).
-  log('\n## Running queries through BEFORE (grep-only) and AFTER (graph + grep)');
-  const results: QueryResult[] = [];
-  for (const q of queries) {
-    // BEFORE: text-fallback only
-    const beforeReturned = beforePrAnswer(q, contentBySlug);
-    let beforeFound = 0;
-    for (const e of q.expected) if (beforeReturned.has(e)) beforeFound++;
-
-    // AFTER (graph-only): traversePaths once per accepted link type, union
-    // results. ("Who works at X?" accepts both works_at and founded — founders
-    // are employees by definition.) Used for the ablation column.
-    const graphOnlyReturned = new Set<string>();
-    for (const lt of q.linkTypes) {
-      const paths = await engine.traversePaths(q.seed, {
-        depth: 1,
-        direction: q.direction,
-        linkType: lt,
-      });
-      for (const p of paths) {
-        const target = q.direction === 'out' ? p.to_slug : p.from_slug;
-        if (target !== q.seed) graphOnlyReturned.add(target);
-      }
-    }
-    let graphOnlyFound = 0;
-    for (const e of q.expected) if (graphOnlyReturned.has(e)) graphOnlyFound++;
-
-    // AFTER (graph-augmented union for SET metrics): graph results union grep.
-    // Same set as grep (graph is a subset of grep in this corpus), preserves
-    // recall identically to BEFORE. Set precision matches grep precision.
-    // The real win is in TOP-K ordering, computed below.
-    const afterReturned = new Set<string>(graphOnlyReturned);
-    for (const r of beforeReturned) afterReturned.add(r);
-    let afterFound = 0;
-    for (const e of q.expected) if (afterReturned.has(e)) afterFound++;
-
-    // Top-K metrics: what the agent actually reads. AFTER ranks graph results
-    // FIRST (high precision), then fills with grep results not in graph.
-    // BEFORE has no ranking signal; we model it as a deterministic but
-    // arbitrary order (the order grep visits pages, which is essentially
-    // alphabetical-by-slug — neutral, no graph influence).
-    const expectedSet = new Set(q.expected);
-
-    const beforeRanked = [...beforeReturned].sort();
-    const beforeTopK = beforeRanked.slice(0, TOP_K);
-    let beforeFoundAtK = 0;
-    for (const r of beforeTopK) if (expectedSet.has(r)) beforeFoundAtK++;
-
-    const graphFirst = [...graphOnlyReturned];
-    const grepRest = [...beforeReturned].filter(r => !graphOnlyReturned.has(r)).sort();
-    const afterRanked = [...graphFirst, ...grepRest];
-    const afterTopK = afterRanked.slice(0, TOP_K);
-    let afterFoundAtK = 0;
-    for (const r of afterTopK) if (expectedSet.has(r)) afterFoundAtK++;
-
-    results.push({
-      question: q.question,
-      expected: q.expected.length,
-      beforeFound,
-      beforeReturned: beforeReturned.size,
-      graphOnlyFound,
-      graphOnlyReturned: graphOnlyReturned.size,
-      afterFound,
-      afterReturned: afterReturned.size,
-      beforeFoundAtK,
-      afterFoundAtK,
-    });
-  }
-
-  await engine.disconnect();
-
-  // ── Aggregate ──
-  const totalExpected = results.reduce((s, r) => s + r.expected, 0);
-  const beforeTotalFound = results.reduce((s, r) => s + r.beforeFound, 0);
-  const beforeTotalReturned = results.reduce((s, r) => s + r.beforeReturned, 0);
-  const graphOnlyTotalFound = results.reduce((s, r) => s + r.graphOnlyFound, 0);
-  const graphOnlyTotalReturned = results.reduce((s, r) => s + r.graphOnlyReturned, 0);
-  const afterTotalFound = results.reduce((s, r) => s + r.afterFound, 0);
-  const afterTotalReturned = results.reduce((s, r) => s + r.afterReturned, 0);
-
-  const beforeRecall = totalExpected > 0 ? beforeTotalFound / totalExpected : 1;
-  const beforePrecision = beforeTotalReturned > 0 ? beforeTotalFound / beforeTotalReturned : 1;
-  const graphOnlyRecall = totalExpected > 0 ? graphOnlyTotalFound / totalExpected : 1;
-  const graphOnlyPrecision = graphOnlyTotalReturned > 0 ? graphOnlyTotalFound / graphOnlyTotalReturned : 1;
-  const afterRecall = totalExpected > 0 ? afterTotalFound / totalExpected : 1;
-  const afterPrecision = afterTotalReturned > 0 ? afterTotalFound / afterTotalReturned : 1;
-
-  // Per-link-type breakdown (group by primary type — first in linkTypes array)
-  const byType: Record<string, { exp: number; bF: number; bR: number; gF: number; gR: number; aF: number; aR: number }> = {};
-  for (let i = 0; i < queries.length; i++) {
-    const q = queries[i];
-    const r = results[i];
-    const t = q.linkTypes[0] ?? 'unknown';
-    byType[t] ??= { exp: 0, bF: 0, bR: 0, gF: 0, gR: 0, aF: 0, aR: 0 };
-    byType[t].exp += r.expected;
-    byType[t].bF += r.beforeFound;
-    byType[t].bR += r.beforeReturned;
-    byType[t].gF += r.graphOnlyFound;
-    byType[t].gR += r.graphOnlyReturned;
-    byType[t].aF += r.afterFound;
-    byType[t].aR += r.afterReturned;
-  }
-
-  // ── Output ──
-  const pct = (v: number) => `${(v * 100).toFixed(1)}%`;
-  const sign = (n: number) => `${n >= 0 ? '+' : ''}${n.toFixed(1)}`;
-  const f1 = (p: number, r: number) => p + r > 0 ? (2 * p * r) / (p + r) : 0;
-
-  const beforeF1 = f1(beforePrecision, beforeRecall);
-  const graphOnlyF1 = f1(graphOnlyPrecision, graphOnlyRecall);
-  const afterF1 = f1(afterPrecision, afterRecall);
-
-  // Top-K aggregates: the metrics that match real agent behavior.
-  const beforeTotalAtK = results.reduce((s, r) => s + r.beforeFoundAtK, 0);
-  const afterTotalAtK = results.reduce((s, r) => s + r.afterFoundAtK, 0);
-  // Each query contributes min(K, returnedSize) to the precision denominator.
-  const beforeReturnedAtK = results.reduce((s, r) => s + Math.min(TOP_K, r.beforeReturned), 0);
-  const afterReturnedAtK = results.reduce((s, r) => s + Math.min(TOP_K, r.afterReturned), 0);
-  const beforePrecAtK = beforeReturnedAtK > 0 ? beforeTotalAtK / beforeReturnedAtK : 0;
-  const afterPrecAtK = afterReturnedAtK > 0 ? afterTotalAtK / afterReturnedAtK : 0;
-  // Recall@K = correct in top-K / total expected.
-  const beforeRecAtK = totalExpected > 0 ? beforeTotalAtK / totalExpected : 0;
-  const afterRecAtK = totalExpected > 0 ? afterTotalAtK / totalExpected : 0;
-
-  // (Earlier benchmark iterations had a "dense queries" slice for queries
-  // where grep returned >> K. Removed — the corpus has small expected counts
-  // per query so the slice was empty. The aggregate top-K already shows the
-  // ranking improvement clearly.)
-
-  log('\n## Headline: top-K relational query accuracy on 240-page rich-prose corpus');
-  log('');
-  log(`Real agents read ranked top-K results, not full sets. AFTER ranks graph hits`);
-  log(`first (high precision) then fills with grep. K=${TOP_K} (a tight ceiling — agents`);
-  log(`almost always read at least the top 5 results).`);
-  log('');
-  log('| Metric                       | BEFORE PR #188 | AFTER PR #188 | Δ                |');
-  log('|------------------------------|----------------|---------------|------------------|');
-  log(`| **Precision@${TOP_K}**                | **${pct(beforePrecAtK)}**         | **${pct(afterPrecAtK)}**        | **${sign((afterPrecAtK - beforePrecAtK) * 100)}pts**         |`);
-  log(`| **Recall@${TOP_K}**                   | **${pct(beforeRecAtK)}**         | **${pct(afterRecAtK)}**        | **${sign((afterRecAtK - beforeRecAtK) * 100)}pts**         |`);
-  log(`| Correct in top-${TOP_K} (total)       | ${String(beforeTotalAtK).padEnd(14)} | ${String(afterTotalAtK).padEnd(13)} | ${sign(afterTotalAtK - beforeTotalAtK).replace('.0','')}              |`);
-  log('');
-  log('## Set-based metrics (full result sets, no top-K cutoff)');
-  log('');
-  log('| Metric                   | BEFORE PR #188 | AFTER PR #188 | Δ              | Graph-only (ablation) |');
-  log('|--------------------------|----------------|---------------|----------------|-----------------------|');
-  log(`| **F1 score**             | **${pct(beforeF1)}**         | **${pct(afterF1)}**        | **${sign((afterF1 - beforeF1) * 100)}pts**       | ${pct(graphOnlyF1).padEnd(21)} |`);
-  log(`| Relational recall        | ${pct(beforeRecall).padEnd(14)} | ${pct(afterRecall).padEnd(13)} | ${sign((afterRecall - beforeRecall) * 100)}pts          | ${pct(graphOnlyRecall).padEnd(21)} |`);
-  log(`| Relational precision     | ${pct(beforePrecision).padEnd(14)} | ${pct(afterPrecision).padEnd(13)} | ${sign((afterPrecision - beforePrecision) * 100)}pts          | ${pct(graphOnlyPrecision).padEnd(21)} |`);
-  log(`| Total returned (any)     | ${String(beforeTotalReturned).padEnd(14)} | ${String(afterTotalReturned).padEnd(13)} | ${sign(afterTotalReturned - beforeTotalReturned).replace('.0','')}             | ${String(graphOnlyTotalReturned).padEnd(21)} |`);
-  log(`| Correct returned         | ${String(beforeTotalFound).padEnd(14)} | ${String(afterTotalFound).padEnd(13)} | ${sign(afterTotalFound - beforeTotalFound).replace('.0','')}              | ${String(graphOnlyTotalFound).padEnd(21)} |`);
-
-  log('\n## By link type (AFTER vs BEFORE, set metrics)');
-  log('| Link type   | Expected | BEFORE found/ret      | AFTER found/ret       | Recall Δ | Precision Δ | F1 Δ        |');
-  log('|-------------|----------|-----------------------|-----------------------|----------|-------------|-------------|');
-  for (const [t, b] of Object.entries(byType)) {
-    const bRec = b.exp > 0 ? b.bF / b.exp : 0;
-    const aRec = b.exp > 0 ? b.aF / b.exp : 0;
-    const bPrec = b.bR > 0 ? b.bF / b.bR : 0;
-    const aPrec = b.aR > 0 ? b.aF / b.aR : 0;
-    const bF1 = f1(bPrec, bRec);
-    const aF1 = f1(aPrec, aRec);
-    log(`| ${t.padEnd(11)} | ${String(b.exp).padEnd(8)} | ${`${b.bF}/${b.bR}`.padEnd(21)} | ${`${b.aF}/${b.aR}`.padEnd(21)} | ${(aRec - bRec >= 0 ? '+' : '')}${((aRec - bRec) * 100).toFixed(0)}pts    | ${(aPrec - bPrec >= 0 ? '+' : '')}${((aPrec - bPrec) * 100).toFixed(0)}pts       | ${(aF1 - bF1 >= 0 ? '+' : '')}${((aF1 - bF1) * 100).toFixed(0)}pts      |`);
-  }
-
-  log('\n## What this proves');
-  log('');
-  log(`PR #188 strictly dominates BEFORE on both top-K metrics — agents see ${afterTotalAtK - beforeTotalAtK}`);
-  log(`more correct answers in their top-${TOP_K} results. Graph hits are surfaced FIRST in`);
-  log(`the ranked list; the agent's first reads are exact-typed answers instead of`);
-  log(`arbitrary text matches. No category goes down.`);
-  log('');
-  log(`Set-based metrics (full result sets) are unchanged because graph hits are a`);
-  log(`subset of grep hits in this corpus — taking the union doesn't add or remove`);
-  log(`anything from the bag of returned results. What changes is which results`);
-  log(`appear FIRST. Top-K captures that; raw set recall doesn't.`);
-  log('');
-  log(`The graph-only ablation column shows the upper bound of where this is going:`);
-  log(`${pct(graphOnlyPrecision)} precision, ${pct(graphOnlyRecall)} recall. The next round of extraction`);
-  log(`tuning (TODOS.md v0.10.5) will lift graph recall toward grep parity, at`);
-  log(`which point set-based metrics also start to favor AFTER.`);
-
-  if (json) {
-    process.stdout.write(JSON.stringify({
-      pages: pages.length,
-      queries: queries.length,
-      before: { recall: beforeRecall, precision: beforePrecision, returned: beforeTotalReturned, found: beforeTotalFound },
-      after: { recall: afterRecall, precision: afterPrecision, returned: afterTotalReturned, found: afterTotalFound },
-      byType,
-      perQuery: results,
-    }, null, 2) + '\n');
-  }
-}
-
-main().catch(e => { console.error(e); process.exit(1); });
diff --git a/eval/runner/identity.ts b/eval/runner/identity.ts
deleted file mode 100644
index a248211d5..000000000
--- a/eval/runner/identity.ts
+++ /dev/null
@@ -1,177 +0,0 @@
-/**
- * BrainBench Category 3: Identity Resolution.
- *
- * Tests whether gbrain can resolve aliases ("Sarah Chen", "S. Chen", "@schen",
- * "sarah.chen@example.com") to one canonical entity.
- *
- * gbrain currently has NO alias table. The benchmark measures what's possible
- * with searchKeyword (tsvector) + slug-based getPage. Numbers will be honest:
- * documented aliases (in canonical body) findable; undocumented not.
- *
- * The point is to surface the gap. A good v1 number on undocumented aliases
- * would mean we have an alias table; a poor number proves we should build one.
- *
- * Usage: bun run eval/runner/identity.ts [--json]
- */
-
-import { PGLiteEngine } from '../../src/core/pglite-engine.ts';
-
-interface Entity {
-  canonicalSlug: string;
-  fullName: string;
-  /** Aliases mentioned IN the canonical page body (should be keyword-findable). */
-  documentedAliases: string[];
-  /** Aliases that exist (handles, emails, typos) but are NOT in any page. */
-  undocumentedAliases: string[];
-}
-
-const FIRST_NAMES = ['Sarah', 'Alice', 'Bob', 'Carol', 'David', 'Eve', 'Frank', 'Grace', 'Henry', 'Iris', 'Jack', 'Kate', 'Liam', 'Mia', 'Noah', 'Olivia', 'Paul', 'Quinn', 'Rachel', 'Sam'];
-const LAST_NAMES = ['Chen', 'Smith', 'Johnson', 'Williams', 'Brown', 'Jones', 'Garcia', 'Miller', 'Davis', 'Rodriguez', 'Martinez', 'Hernandez', 'Lopez', 'Gonzalez', 'Wilson', 'Anderson', 'Thomas', 'Taylor', 'Moore', 'Jackson'];
-const COMPANIES = ['stripe.com', 'acme.io', 'beta.co', 'gamma.dev', 'delta.ai'];
-
-function generateEntities(n: number): Entity[] {
-  const entities: Entity[] = [];
-  for (let i = 0; i < n; i++) {
-    const first = FIRST_NAMES[i % FIRST_NAMES.length];
-    const last = LAST_NAMES[Math.floor(i / FIRST_NAMES.length) % LAST_NAMES.length];
-    const fullName = `${first} ${last}`;
-    const handle = `@${first[0].toLowerCase()}${last.toLowerCase()}`;
-    const email = `${first.toLowerCase()}.${last.toLowerCase()}@${COMPANIES[i % COMPANIES.length]}`;
-    const initial = `${first[0]}. ${last}`;
-    const noSpace = `${first[0]} ${last}`;
-    const typo1 = `${first.slice(0, -1)}${first[first.length - 1]}${first[first.length - 1]} ${last}`;
-    const typo2 = `${first} ${last}n`;
-    const handlePlain = handle.slice(1);
-
-    entities.push({
-      canonicalSlug: `people/${first.toLowerCase()}-${last.toLowerCase()}-${i}`,
-      fullName,
-      documentedAliases: [fullName, handle, email],
-      undocumentedAliases: [initial, noSpace, typo1, typo2, handlePlain],
-    });
-  }
-  return entities;
-}
-
-interface QueryResult {
-  alias: string;
-  canonicalSlug: string;
-  category: 'documented' | 'undocumented';
-  found: boolean;
-  rankPosition: number; // 1-indexed; 0 = not in top-10
-}
-
-async function main() {
-  const json = process.argv.includes('--json');
-  const log = json ? () => {} : console.log;
-
-  log('# BrainBench Category 3: Identity Resolution\n');
-  log(`Generated: ${new Date().toISOString().slice(0, 19)}`);
-
-  const entities = generateEntities(100);
-  log(`Entities: ${entities.length}`);
-  log(`Aliases per entity: ${entities[0].documentedAliases.length} documented + ${entities[0].undocumentedAliases.length} undocumented = ${entities[0].documentedAliases.length + entities[0].undocumentedAliases.length} total`);
-
-  const engine = new PGLiteEngine();
-  await engine.connect({});
-  await engine.initSchema();
-
-  // Seed canonical pages. Each page mentions the entity by full name + handle + email.
-  for (const e of entities) {
-    await engine.putPage(e.canonicalSlug, {
-      type: 'person',
-      title: e.fullName,
-      compiled_truth: `${e.fullName} (also known as ${e.documentedAliases.slice(1).join(', ')}) is a person in our network. Reach them at ${e.documentedAliases[2]}.`,
-      timeline: '',
-    });
-    // Also chunk for searchKeyword.
-    await engine.upsertChunks(e.canonicalSlug, [
-      { chunk_index: 0, chunk_text: `${e.fullName} ${e.documentedAliases.join(' ')}`, chunk_source: 'compiled_truth' },
-    ]);
-  }
-
-  // Run queries.
-  const results: QueryResult[] = [];
-  for (const e of entities) {
-    for (const cat of ['documented', 'undocumented'] as const) {
-      const aliases = cat === 'documented' ? e.documentedAliases : e.undocumentedAliases;
-      for (const alias of aliases) {
-        const r = await engine.searchKeyword(alias, { limit: 10 });
-        // Page-level dedup, keep highest score per slug
-        const seen = new Set<string>();
-        const pages = r.filter(x => { if (seen.has(x.slug)) return false; seen.add(x.slug); return true; });
-        const idx = pages.findIndex(x => x.slug === e.canonicalSlug);
-        results.push({
-          alias,
-          canonicalSlug: e.canonicalSlug,
-          category: cat,
-          found: idx >= 0,
-          rankPosition: idx + 1, // 0 if not found
-        });
-      }
-    }
-  }
-
-  await engine.disconnect();
-
-  // ── Metrics ──
-  const documented = results.filter(r => r.category === 'documented');
-  const undocumented = results.filter(r => r.category === 'undocumented');
-  const docRecall = documented.filter(r => r.found).length / documented.length;
-  const undocRecall = undocumented.filter(r => r.found).length / undocumented.length;
-  const docMrr = documented.reduce((s, r) => s + (r.found ? 1 / r.rankPosition : 0), 0) / documented.length;
-  const undocMrr = undocumented.reduce((s, r) => s + (r.found ? 1 / r.rankPosition : 0), 0) / undocumented.length;
-
-  log('\n## Metrics');
-  log('| Alias category   | Recall (top-10) | MRR    |');
-  log('|------------------|-----------------|--------|');
-  log(`| Documented       | ${(docRecall * 100).toFixed(1)}%             | ${docMrr.toFixed(3)}  |`);
-  log(`| Undocumented     | ${(undocRecall * 100).toFixed(1)}%             | ${undocMrr.toFixed(3)}  |`);
-
-  log('\n## Per-alias-type breakdown (documented)');
-  const docByType: Record<string, { found: number; total: number }> = {};
-  for (const r of documented) {
-    const type = r.alias.startsWith('@') ? 'handle' : r.alias.includes('@') ? 'email' : 'fullname';
-    docByType[type] ??= { found: 0, total: 0 };
-    docByType[type].total++;
-    if (r.found) docByType[type].found++;
-  }
-  for (const [type, { found, total }] of Object.entries(docByType)) {
-    log(`  ${type.padEnd(10)} ${found}/${total} = ${((found / total) * 100).toFixed(1)}%`);
-  }
-
-  log('\n## Per-alias-type breakdown (undocumented)');
-  const undocByType: Record<string, { found: number; total: number }> = {};
-  for (const r of undocumented) {
-    let type: string;
-    if (r.alias.match(/^[A-Z]\. /)) type = 'initial';
-    else if (r.alias.match(/^[A-Z] /)) type = 'no-period';
-    else if (r.alias.match(/[A-Z][a-z]+n$/)) type = 'typo';
-    else if (r.alias.match(/[a-z][a-z]+ /)) type = 'typo';
-    else type = 'handle-plain';
-    undocByType[type] ??= { found: 0, total: 0 };
-    undocByType[type].total++;
-    if (r.found) undocByType[type].found++;
-  }
-  for (const [type, { found, total }] of Object.entries(undocByType)) {
-    log(`  ${type.padEnd(13)} ${found}/${total} = ${((found / total) * 100).toFixed(1)}%`);
-  }
-
-  log('\n## Interpretation');
-  log('Documented aliases (full name, handle, email mentioned in canonical body):');
-  log(`  Recall ${(docRecall * 100).toFixed(1)}% — what current gbrain can do via tsvector keyword match.`);
-  log('Undocumented aliases (initials, typos, handle without @):');
-  log(`  Recall ${(undocRecall * 100).toFixed(1)}% — what current gbrain CAN'T do without an alias table.`);
-  log('');
-  log('Gap: gbrain has no alias table, no fuzzy match, no nickname dictionary.');
-  log('Suggested v0.11 feature: explicit aliases + Levenshtein/phonetic match.');
-
-  if (json) {
-    process.stdout.write(JSON.stringify({
-      results,
-      summary: { docRecall, undocRecall, docMrr, undocMrr, docByType, undocByType },
-    }, null, 2) + '\n');
-  }
-}
-
-main().catch(e => { console.error(e); process.exit(1); });
diff --git a/eval/runner/mcp-contract.ts b/eval/runner/mcp-contract.ts
deleted file mode 100644
index 1b3f99c01..000000000
--- a/eval/runner/mcp-contract.ts
+++ /dev/null
@@ -1,249 +0,0 @@
-/**
- * BrainBench Category 12: MCP Operation Contract.
- *
- * Tests gbrain operation handlers under (trusted local, untrusted remote) ×
- * (valid, boundary, invalid, injection, resource-exhaustion) inputs.
- *
- * Focused on security-boundary operations and limit enforcement. The unit
- * test suite covers happy-path correctness for every op; this benchmark
- * focuses on the contract surface that an attacker probes.
- *
- * Pass criteria:
- *   - Valid input → correct response
- *   - Invalid input → rejected with clear error (not silent corruption)
- *   - Injection attempts → blocked (no SQL injection, no path traversal)
- *   - Trust-boundary differences enforced (ctx.remote=true tighter than false)
- *   - Limit caps enforced (depth, list_pages limit, etc.)
- *
- * Usage: bun run eval/runner/mcp-contract.ts [--json]
- */
-
-import { PGLiteEngine } from '../../src/core/pglite-engine.ts';
-import { operations as OPERATIONS } from '../../src/core/operations.ts';
-import type { OperationContext } from '../../src/core/operations.ts';
-import type { GBrainConfig } from '../../src/core/config.ts';
-
-interface TestResult {
-  name: string;
-  pass: boolean;
-  detail: string;
-}
-
-async function setup(): Promise<{ engine: PGLiteEngine; cleanup: () => Promise<void> }> {
-  const engine = new PGLiteEngine();
-  await engine.connect({});
-  await engine.initSchema();
-  // Seed a small graph for traversal tests.
-  for (let i = 0; i < 10; i++) {
-    await engine.putPage(`people/p${i}`, {
-      type: 'person', title: `P${i}`, compiled_truth: `Person ${i}.`, timeline: '',
-    });
-  }
-  for (let i = 0; i < 10; i++) {
-    await engine.addLink(`people/p${i}`, `people/p${(i + 1) % 10}`, '', 'mentions');
-  }
-  return {
-    engine,
-    cleanup: async () => { await engine.disconnect(); },
-  };
-}
-
-function ctx(remote: boolean, engine: PGLiteEngine): OperationContext {
-  const config: GBrainConfig = { engine: 'pglite', database_path: ':memory:' };
-  const logger = { info: () => {}, warn: () => {}, error: () => {}, debug: () => {} };
-  return { engine, config, logger: logger as never, dryRun: false, remote };
-}
-
-async function runOp(opName: string, params: Record<string, unknown>, c: OperationContext): Promise<{ ok: true; result: unknown } | { ok: false; error: string }> {
-  const op = OPERATIONS.find(o => o.name === opName);
-  if (!op) return { ok: false, error: `unknown operation: ${opName}` };
-  try {
-    const result = await op.handler(c, params);
-    return { ok: true, result };
-  } catch (e) {
-    return { ok: false, error: e instanceof Error ? e.message : String(e) };
-  }
-}
-
-async function main() {
-  const json = process.argv.includes('--json');
-  const log = json ? () => {} : console.log;
-
-  log('# BrainBench Category 12: MCP Operation Contract\n');
-  log(`Generated: ${new Date().toISOString().slice(0, 19)}`);
-  log(`Operations available: ${OPERATIONS.length}`);
-
-  const { engine, cleanup } = await setup();
-  const results: TestResult[] = [];
-
-  // ── Trust boundary: traverse_graph depth cap ──
-  // v0.10.3 hard-caps depth at 10 for remote callers (DoS prevention).
-  log('\n## Trust boundary: traverse_graph depth cap');
-  {
-    const r = await runOp('traverse_graph', { slug: 'people/p0', depth: 1000 }, ctx(true, engine));
-    const pass = r.ok || r.error.includes('depth') || r.error.includes('limit');
-    results.push({
-      name: 'traverse_graph depth=1000 from remote should be capped or rejected',
-      pass,
-      detail: r.ok ? 'capped silently (acceptable)' : `rejected: ${r.error}`,
-    });
-    log(`  ${pass ? '✓' : '✗'} ${results[results.length - 1].name}`);
-  }
-  {
-    const r = await runOp('traverse_graph', { slug: 'people/p0', depth: 5 }, ctx(true, engine));
-    const pass = r.ok;
-    results.push({
-      name: 'traverse_graph depth=5 from remote should succeed (under cap)',
-      pass,
-      detail: r.ok ? 'ok' : `unexpected error: ${r.error}`,
-    });
-    log(`  ${pass ? '✓' : '✗'} ${results[results.length - 1].name}`);
-  }
-
-  // ── Trust boundary: list_pages limit cap ──
-  log('\n## Trust boundary: list_pages limit cap');
-  {
-    const r = await runOp('list_pages', { limit: 1_000_000 }, ctx(true, engine));
-    if (r.ok) {
-      const list = r.result as Array<unknown>;
-      const pass = list.length <= 1000;
-      results.push({
-        name: 'list_pages limit=1M from remote should be clamped',
-        pass,
-        detail: `returned ${list.length} pages (cap should be <= 1000)`,
-      });
-      log(`  ${pass ? '✓' : '✗'} returned ${list.length} pages`);
-    } else {
-      results.push({ name: 'list_pages limit=1M from remote', pass: false, detail: `errored: ${r.error}` });
-      log(`  ✗ unexpected error: ${r.error}`);
-    }
-  }
-
-  // ── Input validation: slug format ──
-  log('\n## Input validation: slug format');
-  {
-    const badSlugs = [
-      { slug: '../etc/passwd', label: 'path traversal' },
-      { slug: '/absolute/path', label: 'absolute path' },
-      { slug: '../../escape', label: 'parent escape' },
-      { slug: 'no-directory', label: 'missing directory prefix' },
-      { slug: '', label: 'empty string' },
-      { slug: 'people/' + 'x'.repeat(10_000), label: 'huge slug (10K chars)' },
-    ];
-    for (const { slug, label } of badSlugs) {
-      const r = await runOp('get_page', { slug }, ctx(true, engine));
-      // Either rejects with error OR returns null (page not found, no crash, no traversal)
-      // The DANGER would be: returns content from outside the brain, or crashes the server.
-      const pass = !r.ok || r.result === null || r.result === undefined;
-      results.push({
-        name: `get_page rejects/null-returns ${label}`,
-        pass,
-        detail: r.ok ? `returned ${typeof r.result}` : `rejected: ${r.error.slice(0, 80)}`,
-      });
-      log(`  ${pass ? '✓' : '✗'} ${label}: ${r.ok ? 'returned ' + typeof r.result : 'rejected'}`);
-    }
-  }
-
-  // ── Input validation: date format ──
-  log('\n## Input validation: date format');
-  {
-    const badDates = ['not-a-date', '2026-13-45', '99999-01-01', '', '../../../etc/passwd'];
-    for (const date of badDates) {
-      const r = await runOp('add_timeline_entry', { slug: 'people/p0', date, summary: 'test' }, ctx(true, engine));
-      // Should reject invalid dates with error.
-      const pass = !r.ok;
-      results.push({
-        name: `add_timeline_entry rejects invalid date "${date.slice(0, 30)}"`,
-        pass,
-        detail: r.ok ? 'silently accepted (BUG)' : `rejected: ${r.error.slice(0, 80)}`,
-      });
-      log(`  ${pass ? '✓' : '✗'} "${date.slice(0, 30)}": ${r.ok ? 'accepted (bad)' : 'rejected'}`);
-    }
-  }
-
-  // ── Injection: SQL injection attempts in search ──
-  log('\n## Injection: SQL injection attempts');
-  {
-    const sqlPayloads = [
-      `'; DROP TABLE pages; --`,
-      `' OR '1'='1`,
-      `'; SELECT * FROM access_tokens; --`,
-      `\\x00\\x00\\x00`,
-      `\u0000injection`,
-    ];
-    for (const payload of sqlPayloads) {
-      const r = await runOp('search', { query: payload, limit: 5 }, ctx(true, engine));
-      // Should NOT throw a SQL parse error (which would mean the input reached SQL).
-      // Should return empty results or a clean error.
-      const pass = r.ok || !r.error.toLowerCase().includes('syntax');
-      results.push({
-        name: `search resists SQL payload "${payload.slice(0, 30)}"`,
-        pass,
-        detail: r.ok ? 'parameterized OK' : `error: ${r.error.slice(0, 80)}`,
-      });
-      log(`  ${pass ? '✓' : '✗'} "${payload.slice(0, 30)}": ${r.ok ? 'safe' : r.error.slice(0, 60)}`);
-    }
-  }
-
-  // ── Resource exhaustion: large inputs ──
-  log('\n## Resource exhaustion: large inputs');
-  {
-    const huge = 'x'.repeat(10_000_000); // 10MB string
-    const start = Date.now();
-    const r = await runOp('search', { query: huge, limit: 5 }, ctx(true, engine));
-    const elapsed = Date.now() - start;
-    const pass = elapsed < 5000; // under 5s
-    results.push({
-      name: 'search with 10MB query string returns within 5s',
-      pass,
-      detail: `${elapsed}ms${r.ok ? ' (returned)' : ` (rejected: ${r.error.slice(0, 60)})`}`,
-    });
-    log(`  ${pass ? '✓' : '✗'} 10MB query: ${elapsed}ms`);
-  }
-
-  // ── Trust boundary: file_upload path confinement ──
-  // Skipped — file_upload requires actual filesystem setup. Covered by unit
-  // tests in test/file-upload-security.test.ts.
-
-  // ── Sanity: every operation has a handler ──
-  log('\n## Sanity: every operation has a handler');
-  for (const op of OPERATIONS) {
-    const pass = typeof op.handler === 'function';
-    results.push({ name: `${op.name} has handler`, pass, detail: pass ? 'ok' : 'missing handler' });
-    if (!pass) log(`  ✗ ${op.name}`);
-  }
-  log(`  ${OPERATIONS.length}/${OPERATIONS.length} operations have handlers`);
-
-  await cleanup();
-
-  // ── Summary ──
-  const passed = results.filter(r => r.pass).length;
-  const failed = results.length - passed;
-
-  log(`\n## Summary`);
-  log(`Tests: ${results.length}`);
-  log(`Passed: ${passed} (${((passed / results.length) * 100).toFixed(1)}%)`);
-  log(`Failed: ${failed}`);
-
-  if (failed > 0) {
-    log('\nFailures:');
-    for (const r of results.filter(r => !r.pass)) {
-      log(`  ✗ ${r.name}`);
-      log(`    ${r.detail}`);
-    }
-  }
-
-  if (json) {
-    process.stdout.write(JSON.stringify({ results, summary: { passed, failed, total: results.length } }, null, 2) + '\n');
-  }
-
-  if (failed > 0) {
-    console.error(`\n⚠ ${failed} contract test(s) failed`);
-    process.exit(1);
-  }
-}
-
-main().catch(e => {
-  console.error('MCP contract eval error:', e);
-  process.exit(1);
-});
diff --git a/eval/runner/perf.ts b/eval/runner/perf.ts
deleted file mode 100644
index 7d22d5f01..000000000
--- a/eval/runner/perf.ts
+++ /dev/null
@@ -1,241 +0,0 @@
-/**
- * BrainBench Category 7: Performance / Latency at scale.
- *
- * Measures gbrain operation latency (P50/P95/P99) and throughput at 1K and 10K
- * page scales. Currently gbrain has zero published performance numbers — this
- * eval changes that.
- *
- * Runs against PGLite (in-memory). A future variant should also run against
- * real Postgres for write-throughput comparison; not in scope today.
- *
- * Usage: bun run eval/runner/perf.ts [--scale 1000|10000] [--json]
- */
-
-import { PGLiteEngine } from '../../src/core/pglite-engine.ts';
-import type { PageInput } from '../../src/core/types.ts';
-
-interface LatencySample {
-  op: string;
-  scale: number;
-  p50_ms: number;
-  p95_ms: number;
-  p99_ms: number;
-  count: number;
-}
-
-interface ThroughputSample {
-  op: string;
-  scale: number;
-  total_seconds: number;
-  ops_per_sec: number;
-}
-
-function percentile(sorted: number[], p: number): number {
-  if (sorted.length === 0) return 0;
-  const idx = Math.min(sorted.length - 1, Math.floor((p / 100) * sorted.length));
-  return sorted[idx];
-}
-
-async function timed<T>(fn: () => Promise<T>): Promise<{ result: T; ms: number }> {
-  const start = performance.now();
-  const result = await fn();
-  return { result, ms: performance.now() - start };
-}
-
-async function timeMany(label: string, scale: number, fn: () => Promise<unknown>, runs: number): Promise<LatencySample> {
-  const samples: number[] = [];
-  for (let i = 0; i < runs; i++) {
-    const { ms } = await timed(fn);
-    samples.push(ms);
-  }
-  samples.sort((a, b) => a - b);
-  return {
-    op: label,
-    scale,
-    p50_ms: percentile(samples, 50),
-    p95_ms: percentile(samples, 95),
-    p99_ms: percentile(samples, 99),
-    count: runs,
-  };
-}
-
-/**
- * Procedural seeder. Power-law connection distribution: 5% of entities are
- * "hub nodes" with many inbound links; the rest are sparsely connected. This
- * matches real-brain shape and stresses the right code paths.
- */
-function generateSeedData(scale: number): { pages: Array<{ slug: string; page: PageInput }>; links: Array<{ from: string; to: string; type: string }> } {
-  const pages: Array<{ slug: string; page: PageInput }> = [];
-  const links: Array<{ from: string; to: string; type: string }> = [];
-
-  // 60% people, 20% companies, 10% meetings, 10% concepts
-  const peopleN = Math.floor(scale * 0.6);
-  const companyN = Math.floor(scale * 0.2);
-  const meetingN = Math.floor(scale * 0.1);
-  const conceptN = Math.floor(scale * 0.1);
-
-  for (let i = 0; i < peopleN; i++) {
-    const slug = `people/person-${i}`;
-    pages.push({
-      slug,
-      page: {
-        type: 'person', title: `Person ${i}`,
-        compiled_truth: `Person ${i} works in tech. Met them via [Company](companies/company-${i % companyN}). Mentioned in [Meeting](meetings/meeting-${i % meetingN}).`,
-        timeline: `- **2025-01-${(i % 28) + 1 < 10 ? '0' : ''}${(i % 28) + 1}** | First met\n- **2025-06-15** | Follow-up call\n- **2026-01-10** | Latest update`,
-      },
-    });
-  }
-  for (let i = 0; i < companyN; i++) {
-    const slug = `companies/company-${i}`;
-    pages.push({
-      slug,
-      page: {
-        type: 'company', title: `Company ${i}`,
-        compiled_truth: `Company ${i} is a startup in fintech.`,
-        timeline: `- **2024-09-01** | Founded\n- **2025-03-15** | Seed round\n- **2026-02-01** | Series A`,
-      },
-    });
-  }
-  for (let i = 0; i < meetingN; i++) {
-    const slug = `meetings/meeting-${i}`;
-    pages.push({
-      slug,
-      page: {
-        type: 'meeting', title: `Meeting ${i}`,
-        compiled_truth: `Meeting ${i} attendees: [Person A](people/person-${i * 5 % peopleN}), [Person B](people/person-${(i * 5 + 1) % peopleN}), [Person C](people/person-${(i * 5 + 2) % peopleN}).`,
-        timeline: `- **2026-03-01** | Meeting held`,
-      },
-    });
-  }
-  for (let i = 0; i < conceptN; i++) {
-    pages.push({
-      slug: `concepts/concept-${i}`,
-      page: {
-        type: 'concept', title: `Concept ${i}`,
-        compiled_truth: `Concept ${i} relates to [Company](companies/company-${i % companyN}).`,
-        timeline: `- **2025-12-01** | Wrote thesis`,
-      },
-    });
-  }
-
-  // Hub-node connections: 5% of people get 100+ inbound links from the rest.
-  const hubCount = Math.max(1, Math.floor(peopleN * 0.05));
-  for (let i = 0; i < hubCount; i++) {
-    const hub = `people/person-${i}`;
-    // Connect every Nth person to this hub.
-    const interval = Math.max(1, Math.floor(peopleN / 100));
-    for (let j = hubCount; j < peopleN; j += interval) {
-      links.push({ from: `people/person-${j}`, to: hub, type: 'mentions' });
-    }
-  }
-
-  return { pages, links };
-}
-
-async function runScale(scale: number, log: (msg: string) => void): Promise<{ latency: LatencySample[]; throughput: ThroughputSample[] }> {
-  log(`\n## Scale: ${scale} pages\n`);
-
-  const engine = new PGLiteEngine();
-  await engine.connect({});
-  await engine.initSchema();
-
-  const { pages, links } = generateSeedData(scale);
-
-  // ── Throughput: bulk import via putPage ──
-  const importStart = performance.now();
-  for (const { slug, page } of pages) {
-    await engine.putPage(slug, page);
-  }
-  const importSecs = (performance.now() - importStart) / 1000;
-  const importTput: ThroughputSample = {
-    op: 'putPage_bulk',
-    scale,
-    total_seconds: importSecs,
-    ops_per_sec: pages.length / importSecs,
-  };
-  log(`Bulk putPage: ${pages.length} pages in ${importSecs.toFixed(1)}s = ${importTput.ops_per_sec.toFixed(1)} pages/sec`);
-
-  // ── Throughput: bulk addLink ──
-  const linkStart = performance.now();
-  for (const l of links) {
-    try { await engine.addLink(l.from, l.to, '', l.type); } catch { /* skip if either page missing */ }
-  }
-  const linkSecs = (performance.now() - linkStart) / 1000;
-  const linkTput: ThroughputSample = {
-    op: 'addLink_bulk',
-    scale,
-    total_seconds: linkSecs,
-    ops_per_sec: links.length / linkSecs,
-  };
-  log(`Bulk addLink: ${links.length} links in ${linkSecs.toFixed(1)}s = ${linkTput.ops_per_sec.toFixed(1)} links/sec`);
-
-  // ── Latency samples ──
-  // Pick 50 random slugs to query.
-  const sampleSlugs: string[] = [];
-  for (let i = 0; i < 50; i++) {
-    sampleSlugs.push(pages[Math.floor(Math.random() * pages.length)].slug);
-  }
-  const hubSlug = `people/person-0`; // known to have many inbound links
-
-  const latency: LatencySample[] = [];
-
-  latency.push(await timeMany('get_page', scale, () => engine.getPage(sampleSlugs[Math.floor(Math.random() * sampleSlugs.length)]), 50));
-  latency.push(await timeMany('get_links', scale, () => engine.getLinks(sampleSlugs[Math.floor(Math.random() * sampleSlugs.length)]), 50));
-  latency.push(await timeMany('get_backlinks', scale, () => engine.getBacklinks(sampleSlugs[Math.floor(Math.random() * sampleSlugs.length)]), 50));
-  latency.push(await timeMany('get_backlinks_hub', scale, () => engine.getBacklinks(hubSlug), 20));
-  latency.push(await timeMany('get_timeline', scale, () => engine.getTimeline(sampleSlugs[Math.floor(Math.random() * sampleSlugs.length)]), 50));
-  latency.push(await timeMany('get_stats', scale, () => engine.getStats(), 10));
-  latency.push(await timeMany('list_pages_50', scale, () => engine.listPages({ limit: 50 }), 20));
-  latency.push(await timeMany('search_keyword', scale, () => engine.searchKeyword('person', { limit: 20 }), 30));
-  latency.push(await timeMany('traverse_paths_d1', scale, () => engine.traversePaths(hubSlug, { depth: 1, direction: 'in' }), 10));
-  latency.push(await timeMany('traverse_paths_d2', scale, () => engine.traversePaths(hubSlug, { depth: 2, direction: 'both' }), 10));
-
-  // Single-page write latency (separate from bulk).
-  let counter = 0;
-  latency.push(await timeMany('putPage_single', scale, () => engine.putPage(`probes/p-${counter++}`, {
-    type: 'concept', title: `P${counter}`, compiled_truth: 'A probe page.', timeline: '',
-  }), 30));
-
-  for (const s of latency) {
-    log(`  ${s.op.padEnd(22)} P50=${s.p50_ms.toFixed(2)}ms  P95=${s.p95_ms.toFixed(2)}ms  P99=${s.p99_ms.toFixed(2)}ms  (n=${s.count})`);
-  }
-
-  await engine.disconnect();
-  return { latency, throughput: [importTput, linkTput] };
-}
-
-async function main() {
-  const json = process.argv.includes('--json');
-  const log = json ? () => {} : console.log;
-
-  const scaleArg = process.argv.findIndex(a => a === '--scale');
-  const scales = scaleArg !== -1 ? [Number(process.argv[scaleArg + 1])] : [1000, 10000];
-
-  log('# BrainBench Category 7: Performance / Latency\n');
-  log(`Generated: ${new Date().toISOString().slice(0, 19)}`);
-  log(`Engine: PGLite (in-memory)`);
-
-  const allLatency: LatencySample[] = [];
-  const allThroughput: ThroughputSample[] = [];
-
-  for (const scale of scales) {
-    const { latency, throughput } = await runScale(scale, log);
-    allLatency.push(...latency);
-    allThroughput.push(...throughput);
-  }
-
-  if (json) {
-    process.stdout.write(JSON.stringify({ latency: allLatency, throughput: allThroughput }, null, 2) + '\n');
-  }
-
-  // Threshold check: P95 search latency at 10K pages should be < 200ms.
-  const search10k = allLatency.find(l => l.op === 'search_keyword' && l.scale === 10000);
-  if (search10k && search10k.p95_ms > 200) {
-    console.error(`\n⚠ search_keyword P95 at 10K = ${search10k.p95_ms.toFixed(1)}ms (threshold 200ms)`);
-  }
-}
-
-main().catch(e => {
-  console.error('Perf benchmark error:', e);
-  process.exit(1);
-});
diff --git a/eval/runner/temporal.ts b/eval/runner/temporal.ts
deleted file mode 100644
index 10acaed55..000000000
--- a/eval/runner/temporal.ts
+++ /dev/null
@@ -1,241 +0,0 @@
-/**
- * BrainBench Category 4: Temporal Queries.
- *
- * Tests:
- *   - Point: "what happened on date X?"
- *   - Range: "what happened between A and B?"
- *   - Recency: "most recent N events for X"
- *   - As-of: "as of date D, where did X work?" (HARD — gbrain has no native op)
- *   - Comparative: "what changed between Q1 and Q2?" (HARD — gbrain has no native op)
- *
- * Compares structured timeline_entries (via getTimeline) against content scan
- * (parsing markdown timeline section in pages.timeline).
- *
- * Usage: bun run eval/runner/temporal.ts [--json]
- */
-
-import { PGLiteEngine } from '../../src/core/pglite-engine.ts';
-
-interface TimelineEvent {
-  slug: string;
-  date: string;
-  summary: string;
-}
-
-interface AsOfQuery {
-  question: string;
-  slug: string;
-  asOfDate: string;
-  /** Expected answer: the most-recent timeline entry summary on or before asOfDate. */
-  expected: string;
-}
-
-function generateData(): { events: TimelineEvent[]; asOfQueries: AsOfQuery[] } {
-  const events: TimelineEvent[] = [];
-  const asOfQueries: AsOfQuery[] = [];
-
-  // 50 entities, each with 10-20 dated events spread over 5 years.
-  for (let i = 0; i < 50; i++) {
-    const slug = `people/p${i}`;
-    const eventCount = 10 + (i % 10);
-    let currentJob = 'startup-0';
-    for (let e = 0; e < eventCount; e++) {
-      // Date: 2021 to 2026, evenly spaced + jittered.
-      const baseDays = (e / eventCount) * (5 * 365);
-      const jitter = (i * 13 + e * 7) % 30;
-      const dayOffset = Math.floor(baseDays + jitter);
-      const d = new Date('2021-01-01');
-      d.setUTCDate(d.getUTCDate() + dayOffset);
-      const date = d.toISOString().slice(0, 10);
-      // Event types: job change, funding, talk, mention.
-      const eventTypes = ['joined', 'announced', 'spoke at', 'hired by', 'promoted to'];
-      const summary = `${eventTypes[e % eventTypes.length]} startup-${e % 5}`;
-      events.push({ slug, date, summary });
-
-      if (eventTypes[e % eventTypes.length] === 'joined' || eventTypes[e % eventTypes.length] === 'hired by') {
-        currentJob = `startup-${e % 5}`;
-      }
-    }
-
-    // As-of query: pick a date in 2024 and ask where this person worked at that time.
-    const asOfDate = `2024-06-${15 + (i % 14)}`;
-    // Expected: the most recent "joined" or "hired by" event before asOfDate.
-    const before = events
-      .filter(ev => ev.slug === slug && (ev.summary.startsWith('joined') || ev.summary.startsWith('hired by')) && ev.date <= asOfDate)
-      .sort((a, b) => b.date.localeCompare(a.date));
-    if (before.length > 0) {
-      asOfQueries.push({
-        question: `As of ${asOfDate}, where did p${i} work?`,
-        slug,
-        asOfDate,
-        expected: before[0].summary,
-      });
-    }
-  }
-
-  return { events, asOfQueries };
-}
-
-async function main() {
-  const json = process.argv.includes('--json');
-  const log = json ? () => {} : console.log;
-
-  log('# BrainBench Category 4: Temporal Queries\n');
-  log(`Generated: ${new Date().toISOString().slice(0, 19)}`);
-
-  const { events, asOfQueries } = generateData();
-  log(`Events: ${events.length}`);
-  log(`Entities: 50`);
-  log(`As-of queries: ${asOfQueries.length}`);
-
-  const engine = new PGLiteEngine();
-  await engine.connect({});
-  await engine.initSchema();
-
-  // Seed pages with timeline content (markdown form) AND structured entries.
-  const eventsBySlug = new Map<string, TimelineEvent[]>();
-  for (const e of events) {
-    if (!eventsBySlug.has(e.slug)) eventsBySlug.set(e.slug, []);
-    eventsBySlug.get(e.slug)!.push(e);
-  }
-
-  for (const [slug, slugEvents] of eventsBySlug) {
-    const timelineMd = slugEvents.map(e => `- **${e.date}** | ${e.summary}`).join('\n');
-    await engine.putPage(slug, {
-      type: 'person',
-      title: slug.replace('people/', ''),
-      compiled_truth: 'A person.',
-      timeline: timelineMd,
-    });
-    // Also add structured entries.
-    for (const e of slugEvents) {
-      await engine.addTimelineEntry(slug, { date: e.date, summary: e.summary, source: '', detail: '' });
-    }
-  }
-
-  // ── Test 1: Point queries ──
-  // "What happened on YYYY-MM-DD?" — pick 30 random dates that have at least one event.
-  log('\n## Point queries');
-  const dates = [...new Set(events.map(e => e.date))].sort();
-  const testDates = dates.filter((_, i) => i % Math.floor(dates.length / 30) === 0).slice(0, 30);
-  let pointHits = 0, pointTotal = 0, pointReturned = 0, pointValid = 0;
-  for (const date of testDates) {
-    // Use getTimeline filtered manually (no native cross-entity date query).
-    const allEntries: { slug: string; summary: string }[] = [];
-    for (const slug of eventsBySlug.keys()) {
-      const tl = await engine.getTimeline(slug);
-      for (const t of tl) {
-        const tDate = t.date instanceof Date ? t.date.toISOString().slice(0, 10) : String(t.date).slice(0, 10);
-        if (tDate === date) allEntries.push({ slug, summary: t.summary });
-      }
-    }
-    const expected = events.filter(e => e.date === date);
-    pointTotal += expected.length;
-    pointReturned += allEntries.length;
-    for (const e of expected) {
-      if (allEntries.some(a => a.slug === e.slug && a.summary === e.summary)) pointHits++;
-    }
-    for (const a of allEntries) {
-      if (expected.some(e => e.slug === a.slug && e.summary === a.summary)) pointValid++;
-    }
-  }
-  const pointRecall = pointTotal > 0 ? pointHits / pointTotal : 1;
-  const pointPrecision = pointReturned > 0 ? pointValid / pointReturned : 1;
-  log(`  ${testDates.length} dates queried, ${pointTotal} expected events`);
-  log(`  Recall: ${(pointRecall * 100).toFixed(1)}%, Precision: ${(pointPrecision * 100).toFixed(1)}%`);
-
-  // ── Test 2: Range queries ──
-  log('\n## Range queries');
-  const ranges = [
-    { from: '2024-01-01', to: '2024-03-31', label: 'Q1 2024' },
-    { from: '2025-04-01', to: '2025-06-30', label: 'Q2 2025' },
-    { from: '2026-01-01', to: '2026-12-31', label: '2026 full year' },
-    { from: '2023-07-01', to: '2023-09-30', label: 'Q3 2023' },
-  ];
-  let rangeRecall = 0, rangePrecision = 0;
-  for (const r of ranges) {
-    const expected = events.filter(e => e.date >= r.from && e.date <= r.to);
-    const allEntries: { slug: string; date: string; summary: string }[] = [];
-    for (const slug of eventsBySlug.keys()) {
-      const tl = await engine.getTimeline(slug);
-      for (const t of tl) {
-        const tDate = t.date instanceof Date ? t.date.toISOString().slice(0, 10) : String(t.date).slice(0, 10);
-        if (tDate >= r.from && tDate <= r.to) allEntries.push({ slug, date: tDate, summary: t.summary });
-      }
-    }
-    const hits = expected.filter(e => allEntries.some(a => a.slug === e.slug && a.summary === e.summary)).length;
-    const valid = allEntries.filter(a => expected.some(e => e.slug === a.slug && e.summary === a.summary)).length;
-    const recall = expected.length > 0 ? hits / expected.length : 1;
-    const precision = allEntries.length > 0 ? valid / allEntries.length : 1;
-    rangeRecall += recall / ranges.length;
-    rangePrecision += precision / ranges.length;
-    log(`  ${r.label}: ${expected.length} expected, ${allEntries.length} returned, R=${(recall * 100).toFixed(1)}%, P=${(precision * 100).toFixed(1)}%`);
-  }
-  log(`  Average: R=${(rangeRecall * 100).toFixed(1)}%, P=${(rangePrecision * 100).toFixed(1)}%`);
-
-  // ── Test 3: Recency queries ──
-  log('\n## Recency queries (most recent 3 events per entity)');
-  let recencyCorrect = 0, recencyTotal = 0;
-  const sampleEntities = [...eventsBySlug.keys()].slice(0, 30);
-  for (const slug of sampleEntities) {
-    const expected = events.filter(e => e.slug === slug).sort((a, b) => b.date.localeCompare(a.date)).slice(0, 3);
-    const tl = await engine.getTimeline(slug);
-    const sortedTl = [...tl].sort((a, b) => {
-      const ad = a.date instanceof Date ? a.date.toISOString() : String(a.date);
-      const bd = b.date instanceof Date ? b.date.toISOString() : String(b.date);
-      return bd.localeCompare(ad);
-    }).slice(0, 3);
-    for (const e of expected) {
-      recencyTotal++;
-      const sd = sortedTl.find(t => {
-        const tDate = t.date instanceof Date ? t.date.toISOString().slice(0, 10) : String(t.date).slice(0, 10);
-        return tDate === e.date && t.summary === e.summary;
-      });
-      if (sd) recencyCorrect++;
-    }
-  }
-  const recencyAcc = recencyTotal > 0 ? recencyCorrect / recencyTotal : 1;
-  log(`  ${sampleEntities.length} entities × 3 most-recent events each`);
-  log(`  Top-3 correctness: ${(recencyAcc * 100).toFixed(1)}%`);
-
-  // ── Test 4: As-of queries ──
-  log('\n## As-of queries (HARD — no native gbrain operation)');
-  log('  Approach: read full timeline, filter events ≤ asOfDate, take most-recent matching entry.');
-  let asOfCorrect = 0;
-  for (const q of asOfQueries) {
-    const tl = await engine.getTimeline(q.slug);
-    const eligible = tl
-      .filter(t => {
-        const tDate = t.date instanceof Date ? t.date.toISOString().slice(0, 10) : String(t.date).slice(0, 10);
-        return tDate <= q.asOfDate && (t.summary.startsWith('joined') || t.summary.startsWith('hired by'));
-      })
-      .sort((a, b) => {
-        const ad = a.date instanceof Date ? a.date.toISOString() : String(a.date);
-        const bd = b.date instanceof Date ? b.date.toISOString() : String(b.date);
-        return bd.localeCompare(ad);
-      });
-    if (eligible.length > 0 && eligible[0].summary === q.expected) asOfCorrect++;
-  }
-  const asOfAcc = asOfQueries.length > 0 ? asOfCorrect / asOfQueries.length : 1;
-  log(`  ${asOfQueries.length} as-of queries, ${asOfCorrect} correct = ${(asOfAcc * 100).toFixed(1)}%`);
-  log('  Note: requires manual filter+sort logic per query. A native `getStateAtTime`');
-  log('  operation would make this trivial. Suggested v0.11 feature.');
-
-  await engine.disconnect();
-
-  log('\n## Summary');
-  log('| Sub-category    | Recall | Precision | Notes                                |');
-  log('|-----------------|--------|-----------|--------------------------------------|');
-  log(`| Point           | ${(pointRecall * 100).toFixed(1)}% | ${(pointPrecision * 100).toFixed(1)}%    | Cross-entity date query (manual)     |`);
-  log(`| Range           | ${(rangeRecall * 100).toFixed(1)}% | ${(rangePrecision * 100).toFixed(1)}%    | Same — manual cross-entity filter    |`);
-  log(`| Recency (top-3) | ${(recencyAcc * 100).toFixed(1)}% | —         | Per-entity, native getTimeline       |`);
-  log(`| As-of           | ${(asOfAcc * 100).toFixed(1)}% | —         | Hard, no native op (filter+sort)     |`);
-
-  if (json) {
-    process.stdout.write(JSON.stringify({
-      summary: { pointRecall, pointPrecision, rangeRecall, rangePrecision, recencyAcc, asOfAcc },
-    }, null, 2) + '\n');
-  }
-}
-
-main().catch(e => { console.error(e); process.exit(1); });
diff --git a/llms-full.txt b/llms-full.txt
index 14734949b..abbe04514 100644
--- a/llms-full.txt
+++ b/llms-full.txt
@@ -123,7 +123,14 @@ strict behavior when unset.
 - `src/core/embedding.ts` — OpenAI text-embedding-3-large, batch, retry, backoff
 - `src/core/check-resolvable.ts` — Resolver validation: reachability, MECE overlap, DRY checks, structured fix objects. v0.14.1: `CROSS_CUTTING_PATTERNS.conventions` is an array (notability gate accepts both `conventions/quality.md` and `_brain-filing-rules.md`). New `extractDelegationTargets()` parses `> **Convention:**`, `> **Filing rule:**`, and inline backtick references. DRY suppression is proximity-based via `DRY_PROXIMITY_LINES = 40`.
 - `src/core/repo-root.ts` — Shared `findRepoRoot(startDir?)` (v0.16.4): walks up from `startDir` (default `process.cwd()`) looking for `skills/RESOLVER.md`. Zero-dependency module imported by both `doctor.ts` and `check-resolvable.ts`. Parameterized `startDir` makes tests hermetic.
-- `src/commands/check-resolvable.ts` — Standalone CLI wrapper (v0.16.4) over `checkResolvable()`. Exports `parseFlags`, `resolveSkillsDir`, `DEFERRED`, `runCheckResolvable`. Exit rule: **1 on any issue (warnings OR errors)**, stricter than doctor's `ok` flag — honors README:259. Stable JSON envelope `{ok, skillsDir, report, autoFix, deferred, error, message}` — same shape on success and error paths. `--fix` path runs `autoFixDryViolations` BEFORE `checkResolvable` (same ordering as doctor). `deferred[]` array surfaces pending Checks 5 (trigger routing eval) and 6 (brain filing) with issue URLs. `scripts/skillify-check.ts` subprocess-calls `gbrain check-resolvable --json` (cached per process) and fails loud on binary-missing — no silent false-pass.
+- `src/commands/check-resolvable.ts` — Standalone CLI wrapper (v0.16.4) over `checkResolvable()`. Exports `parseFlags`, `resolveSkillsDir`, `DEFERRED`, `runCheckResolvable`. Exit rule: **1 on any issue (warnings OR errors)**, stricter than doctor's `ok` flag — honors README:259. Stable JSON envelope `{ok, skillsDir, report, autoFix, deferred, error, message}` — same shape on success and error paths. `--fix` path runs `autoFixDryViolations` BEFORE `checkResolvable` (same ordering as doctor). `scripts/skillify-check.ts` subprocess-calls `gbrain check-resolvable --json` (cached per process) and fails loud on binary-missing — no silent false-pass. **v0.19:** AGENTS.md workspaces now resolve natively (see `src/core/resolver-filenames.ts`) — gbrain inspects the 107-skill OpenClaw deployment whether the routing file is `RESOLVER.md` or `AGENTS.md`. `DEFERRED[]` is empty — Checks 5 + 6 shipped as real code, not issue URLs.
+- `src/core/resolver-filenames.ts` (v0.19) — central list of accepted routing filenames (`RESOLVER.md`, `AGENTS.md`). Shared by `findRepoRoot`, `check-resolvable`, and skillpack install so every code path walks the same fallback chain.
+- `src/commands/skillify.ts` + `src/core/skillify/{generator,templates}.ts` (v0.19) — `gbrain skillify scaffold <name>` creates all stubs for a new skill in one command: SKILL.md, script, tests, routing-eval.jsonl, resolver entry, filing-rules pointer. `gbrain skillify check <script>` runs the 10-step checklist (LLM evals, routing evals, check-resolvable gate, filing audit) against a candidate skill before it lands.
+- `src/commands/skillify-check.ts` (v0.19) — `gbrain skillpack-check` agent-readable health report. Exit 0/1/2 for CI pipeline gating; JSON for debugging. Wraps `check-resolvable --json`, `doctor --json`, and migration ledger into one payload so agents can decide whether a human action is required.
+- `src/commands/skillpack.ts` + `src/core/skillpack/{bundle,installer}.ts` (v0.19) — `gbrain skillpack install` drops gbrain's curated 25-skill bundle into a host workspace, managed-block style. Never clobbers local edits; tracks a skill manifest so subsequent `install --update` diffs cleanly. Bundle builder (`skillpack/bundle.ts`) packages the set from `skills/` into a versioned payload.
+- `src/core/skill-manifest.ts` (v0.19) — parser for `skill-manifest.json` records. Used by skillpack installer to detect drift between the shipped bundle and the user's local edits, so updates merge instead of overwriting.
+- `src/commands/routing-eval.ts` + `src/core/routing-eval.ts` (v0.19) — `gbrain routing-eval` catches user phrasings that route to the wrong skill. Reads `skills/<name>/routing-eval.jsonl` fixtures (`{intent, expected_skill, ambiguous_with?}`). Structural layer runs in `check-resolvable` by default (zero API cost); `--llm` opts into a Haiku tie-break layer for CI. False positives surface before users hit them.
+- `src/core/filing-audit.ts` + `skills/_brain-filing-rules.json` (v0.19) — Check 6 of `check-resolvable`. Parses new `writes_pages:` / `writes_to:` frontmatter on skills and audits their filing claims against the filing-rules JSON. Warning-only in v0.19, upgrades to error in v0.20.
 - `src/core/dry-fix.ts` — `gbrain doctor --fix` engine. `autoFixDryViolations(fixes, {dryRun})` rewrites inlined rules to `> **Convention:** see [path](path).` callouts via three shape-aware expanders (bullet / blockquote / paragraph). Five guards: working-tree-dirty (`getWorkingTreeStatus()` returns 3-state `'clean' | 'dirty' | 'not_a_repo'`), no-git-backup, inside-code-fence, already-delegated (40-line proximity, consistent with detector), ambiguous-multi-match, block-is-callout. `execFileSync` array args (no shell — no injection surface). EOF newline preserved.
 - `src/core/backoff.ts` — Adaptive load-aware throttling: CPU/memory checks, exponential backoff, active hours multiplier
 - `src/core/fail-improve.ts` — Deterministic-first, LLM-fallback loop with JSONL failure logging and auto-test generation
@@ -192,7 +199,7 @@ strict behavior when unset.
 - `docs/guides/diligence-ingestion.md` — Data room to brain pages pipeline
 - `docs/designs/HOMEBREW_FOR_PERSONAL_AI.md` — 10-star vision for integration system
 - `docs/mcp/` — Per-client setup guides (Claude Desktop, Code, Cowork, Perplexity)
-- `docs/benchmarks/` — Search quality benchmark results (reproducible, fictional data)
+- BrainBench (benchmark suite + corpus): lives in the separate [gbrain-evals](https://github.com/garrytan/gbrain-evals) repo. Not installed alongside gbrain.
 - `skills/_brain-filing-rules.md` — Cross-cutting brain filing rules (referenced by all brain-writing skills)
 - `skills/RESOLVER.md` — Skill routing table (based on the agent-fork AGENTS.md pattern)
 - `skills/conventions/` — Cross-cutting rules (quality, brain-first, model-routing, test-before-bulk, cross-modal)
@@ -223,6 +230,21 @@ strict behavior when unset.
 - `src/commands/report.ts` — Structured report saver (audit trail for maintenance/enrichment)
 - `openclaw.plugin.json` — ClawHub bundle plugin manifest
 
+### BrainBench — in a sibling repo (v0.20+)
+
+BrainBench — the public benchmark for personal-knowledge agent stacks — lives in
+[github.com/garrytan/gbrain-evals](https://github.com/garrytan/gbrain-evals). It
+depends on gbrain as a consumer; gbrain never pulls in the ~5MB eval corpus or
+the pdf-parse dev dep at install time.
+
+gbrain's public API surface (the exports map in `package.json`) is what
+gbrain-evals consumes: `gbrain/engine`, `gbrain/types`, `gbrain/operations`,
+`gbrain/pglite-engine`, `gbrain/link-extraction`, `gbrain/import-file`,
+`gbrain/transcription`, `gbrain/embedding`, `gbrain/config`, `gbrain/markdown`,
+`gbrain/backoff`, `gbrain/search/hybrid`, `gbrain/search/expansion`,
+`gbrain/extract`. Removing any of these is a breaking change for the
+gbrain-evals consumer.
+
 ## Commands
 
 Run `gbrain --help` or `gbrain --tools-json` for full command reference.
@@ -315,7 +337,15 @@ parity), `test/cli.test.ts` (CLI structure), `test/config.test.ts` (config redac
 `test/sync.test.ts` (sync logic + v0.12.3 regression guard asserting top-level `engine.transaction` is not called),
 `test/doctor.test.ts` (doctor command + v0.12.3 assertions that `jsonb_integrity` scans the four v0.12.0 write sites and `markdown_body_completeness` is present),
 `test/utils.test.ts` (shared SQL utilities + `tryParseEmbedding` null-return and single-warn semantics),
-`test/build-llms.test.ts` (llms.txt/llms-full.txt generator: path resolution, idempotence, spec shape, regen-drift guard, content contract, AGENTS.md install-path mirror, size-budget enforcement — 7 cases).
+`test/build-llms.test.ts` (llms.txt/llms-full.txt generator: path resolution, idempotence, spec shape, regen-drift guard, content contract, AGENTS.md install-path mirror, size-budget enforcement — 7 cases),
+`test/check-resolvable-cli.test.ts` (v0.19 CLI wrapper: exit codes, JSON envelope shape, AGENTS.md fallback chain),
+`test/regression-v0_16_4.test.ts` (findRepoRoot regression guard — hermetic startDir parameterization),
+`test/filing-audit.test.ts` (v0.19 Check 6: `writes_pages` / `writes_to` frontmatter, filing-rules JSON validation),
+`test/routing-eval.test.ts` (v0.19 Check 5: fixture parsing, structural routing, ambiguous_with, Haiku tie-break layer),
+`test/skill-manifest.test.ts` (v0.19 skill manifest parser: drift detection, managed-block markers),
+`test/skillify-scaffold.test.ts` (v0.19 `gbrain skillify scaffold` stubs: SKILL.md, script, tests, routing-eval fixtures),
+`test/skillpack-install.test.ts` (v0.19 `gbrain skillpack install` managed-block install / update / no-clobber semantics),
+`test/skillpack-sync-guard.test.ts` (v0.19 sync-guard: bundled skills stay byte-identical to `skills/` source).
 
 E2E tests (`test/e2e/`): Run against real Postgres+pgvector. Require `DATABASE_URL`.
 - `bun run test:e2e` runs Tier 1 (mechanical, all operations, no API keys). Includes 9 dedicated cases for the postgres-engine `addLinksBatch` / `addTimelineEntriesBatch` bind path — postgres-js's `unnest()` binding is structurally different from PGLite's and gets its own coverage.
@@ -324,6 +354,7 @@ E2E tests (`test/e2e/`): Run against real Postgres+pgvector. Require `DATABASE_U
 - `test/e2e/postgres-jsonb.test.ts` — v0.12.2 regression test. Round-trips all 5 JSONB write sites (pages.frontmatter, raw_data.data, ingest_log.pages_updated, files.metadata, page_versions.frontmatter) against real Postgres and asserts `jsonb_typeof='object'` plus `->>'key'` returns the expected scalar. The test that should have caught the original double-encode bug.
 - `test/e2e/jsonb-roundtrip.test.ts` — v0.12.3 companion regression against the 4 doctor-scanned JSONB sites. Assertion-level overlap with `postgres-jsonb.test.ts` is intentional defense-in-depth: if doctor's scan surface ever drifts from the actual write surface, one of these tests catches it.
 - `test/e2e/upgrade.test.ts` runs check-update E2E against real GitHub API (network required)
+- `test/e2e/openclaw-reference-compat.test.ts` (v0.19) — exercises `check-resolvable` + `skillpack install` against a minimal AGENTS.md workspace fixture (`test/fixtures/openclaw-reference-minimal/`), regression guard for the 107-skill OpenClaw deployment shape
 - Tier 2 (`skills.test.ts`) requires OpenClaw + API keys, runs nightly in CI
 - If `.env.testing` doesn't exist in this directory, check sibling worktrees for one:
   `find ../  -maxdepth 2 -name .env.testing -print -quit` and copy it here if found.
@@ -544,7 +575,7 @@ Voice rules:
 
 Source material to pull from:
 - CHANGELOG.md previous entry for prior context
-- `docs/benchmarks/[latest].md` for the headline numbers
+- Latest `gbrain-evals/docs/benchmarks/[latest].md` for headline numbers (sibling repo)
 - Recent commits (`git log <prev-version>..HEAD --oneline`) for what shipped
 - Don't make up numbers. If a metric isn't in a benchmark or production data, don't
   include it. Say "no measurement yet" if asked.
@@ -1165,7 +1196,7 @@ Your AI agent is smart but forgetful. GBrain gives it a brain.
 
 Built by the President and CEO of Y Combinator to run his actual AI agents. The production brain powering his OpenClaw and Hermes deployments: **17,888 pages, 4,383 people, 723 companies**, 21 cron jobs running autonomously, built in 12 days. The agent ingests meetings, emails, tweets, voice calls, and original ideas while you sleep. It enriches every person and company it encounters. It fixes its own citations and consolidates memory overnight. You wake up and the brain is smarter than when you went to bed.
 
-The brain wires itself. Every page write extracts entity references and creates typed links (`attended`, `works_at`, `invested_in`, `founded`, `advises`) with zero LLM calls. Hybrid search. Self-wiring knowledge graph. Structured timeline. Backlink-boosted ranking. Ask "who works at Acme AI?" or "what did Bob invest in this quarter?" and get answers vector search alone can't reach. Benchmarked end-to-end: **Recall@5 jumps from 83% to 95%, Precision@5 from 39% to 45%, +30 more correct answers in the agent's top-5 reads** on a 240-page Opus-generated rich-prose corpus. Graph-only F1: **86.6% vs grep's 57.8%** (+28.8 pts). [Full report](docs/benchmarks/2026-04-18-brainbench-v1.md).
+The brain wires itself. Every page write extracts entity references and creates typed links (`attended`, `works_at`, `invested_in`, `founded`, `advises`) with zero LLM calls. Hybrid search. Self-wiring knowledge graph. Structured timeline. Backlink-boosted ranking. Ask "who works at Acme AI?" or "what did Bob invest in this quarter?" and get answers vector search alone can't reach. Benchmarked side-by-side against the category: gbrain lands **P@5 49.1%, R@5 97.9%** on a 240-page Opus-generated rich-prose corpus, beating its own graph-disabled variant by **+31.4 points P@5** and ripgrep-BM25 + vector-only RAG by a similar margin. The graph layer plus v0.12 extract quality together carry the gap. Full BrainBench scorecards + corpus live in the sibling [gbrain-evals](https://github.com/garrytan/gbrain-evals) repo.
 
 GBrain is those patterns, generalized. 28 skills. Install in 30 minutes. Your agent does the work. As Garry's personal agent gets smarter, so does yours.
 
@@ -1355,7 +1386,7 @@ Here's my personal OpenClaw deployment: one Render container. Supabase Postgres
 
 Under that 19-cron load, sub-agent spawn couldn't clear the 10-second gateway wall. Minions landed it in under a second for zero tokens. **Scaling:** 19,240 posts across 36 months, single bash loop, ~15 min total, $0.00. Sub-agents: ~9 min best case, ~$1.08 in tokens, ~40% spawn failure. **Lab:** durability ∞ (SIGKILL mid-flight, 10/10 rescued), throughput ~10× faster, fan-out ~21× with no failure wall, memory ~400× less.
 
-Full benchmarks: [production](docs/benchmarks/2026-04-18-minions-vs-openclaw-production.md) and [lab](docs/benchmarks/2026-04-18-minions-vs-openclaw-subagents.md).
+Full benchmarks live in [gbrain-evals](https://github.com/garrytan/gbrain-evals/tree/main/docs/benchmarks).
 
 ### The routing rule
 
@@ -1599,7 +1630,7 @@ gbrain extract links --source db        # wire up the existing 29K pages
 gbrain extract timeline --source db     # extract dated events from markdown timelines
 ```
 
-Then ask graph questions or watch the search ranking improve. Benchmarked: **Recall@5 jumps from 83% to 95%, Precision@5 from 39% to 45%, +30 more correct answers in the agent's top-5 reads** on a 240-page Opus-generated rich-prose corpus. Graph-only F1 hits 86.6% vs grep's 57.8% (+28.8 pts). See [docs/benchmarks/2026-04-18-brainbench-v1.md](docs/benchmarks/2026-04-18-brainbench-v1.md).
+Then ask graph questions or watch the search ranking improve. Benchmarked side-by-side against ripgrep-BM25, vector-only RAG (same embedder), and gbrain-with-graph-disabled: gbrain lands **P@5 49.1%, R@5 97.9%** on a 240-page Opus-generated rich-prose corpus, beating hybrid-nograph by **+31.4 points P@5**. Isolate the contribution: v0.11→v0.12 moved the same gbrain codebase from P@5 22.1% → 49.1% on identical inputs, so typed-link extract quality is load-bearing. Full scorecards + reproducible corpus: [gbrain-evals](https://github.com/garrytan/gbrain-evals).
 
 ## Search
 
@@ -1675,7 +1706,7 @@ End-to-end on the BrainBench v1 corpus (240 rich-prose pages, before/after PR #1
 
 Plus 5 orthogonal capability checks (identity resolution, temporal queries,
 performance at 10K-page scale, robustness to malformed input, MCP operation
-contract). All pass. [Full report.](docs/benchmarks/2026-04-18-brainbench-v1.md)
+contract). All pass. Full report: [gbrain-evals](https://github.com/garrytan/gbrain-evals).
 
 The point: each technique handles a class of inputs the others miss. Vector
 search misses exact slug refs; keyword catches them. Keyword misses conceptual
@@ -1837,7 +1868,7 @@ The skills in this repo are those patterns, generalized. What took 11 days to bu
 - [CHANGELOG.md](CHANGELOG.md) ... Version history
 
 **Benchmarks:**
-- [BrainBench v1 (PR #188)](docs/benchmarks/2026-04-18-brainbench-v1.md) ... single comprehensive before/after report on a 240-page Opus-generated corpus. 7 categories: relational queries, identity resolution, temporal queries, performance, robustness, MCP contract.
+- [gbrain-evals](https://github.com/garrytan/gbrain-evals) ... BrainBench, the sibling repo that holds the eval harness, corpus, scorecards, and 4-adapter comparisons. Depends on gbrain; not installed alongside gbrain.
 
 ## Contributing
 
diff --git a/llms.txt b/llms.txt
index 8342207f2..2b8c6bf18 100644
--- a/llms.txt
+++ b/llms.txt
@@ -41,7 +41,6 @@ Repo: https://github.com/garrytan/gbrain
 
 ## Optional
 
-- [docs/benchmarks/](https://raw.githubusercontent.com/garrytan/gbrain/master/docs/benchmarks/): Retrieval quality benchmarks.
 - [docs/designs/](https://raw.githubusercontent.com/garrytan/gbrain/master/docs/designs/): Forward-looking designs.
 - [docs/architecture/infra-layer.md](https://raw.githubusercontent.com/garrytan/gbrain/master/docs/architecture/infra-layer.md): Shared infra patterns.
 
diff --git a/package.json b/package.json
index f7582af14..7011e391d 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "gbrain",
-  "version": "0.19.1",
+  "version": "0.20.0",
   "description": "Postgres-native personal knowledge brain with hybrid RAG search",
   "type": "module",
   "main": "src/core/index.ts",
@@ -13,7 +13,18 @@
     "./types": "./src/core/types.ts",
     "./operations": "./src/core/operations.ts",
     "./minions": "./src/core/minions/index.ts",
-    "./engine-factory": "./src/core/engine-factory.ts"
+    "./engine-factory": "./src/core/engine-factory.ts",
+    "./pglite-engine": "./src/core/pglite-engine.ts",
+    "./link-extraction": "./src/core/link-extraction.ts",
+    "./import-file": "./src/core/import-file.ts",
+    "./transcription": "./src/core/transcription.ts",
+    "./embedding": "./src/core/embedding.ts",
+    "./config": "./src/core/config.ts",
+    "./markdown": "./src/core/markdown.ts",
+    "./backoff": "./src/core/backoff.ts",
+    "./search/hybrid": "./src/core/search/hybrid.ts",
+    "./search/expansion": "./src/core/search/expansion.ts",
+    "./extract": "./src/commands/extract.ts"
   },
   "scripts": {
     "dev": "bun run src/cli.ts",
diff --git a/scripts/llms-config.ts b/scripts/llms-config.ts
index 08dd4449c..9f66af262 100644
--- a/scripts/llms-config.ts
+++ b/scripts/llms-config.ts
@@ -177,12 +177,6 @@ export const SECTIONS: DocSection[] = [
     heading: "Optional",
     optional: true,
     entries: [
-      {
-        title: "docs/benchmarks/",
-        description: "Retrieval quality benchmarks.",
-        path: "docs/benchmarks/",
-        includeInFull: false,
-      },
       {
         title: "docs/designs/",
         description: "Forward-looking designs.",
diff --git a/src/commands/migrations/v0_12_0.ts b/src/commands/migrations/v0_12_0.ts
index a839d4416..0bbe82d3f 100644
--- a/src/commands/migrations/v0_12_0.ts
+++ b/src/commands/migrations/v0_12_0.ts
@@ -244,7 +244,7 @@ export const v0_12_0: Migration = {
       '"what did Bob invest in?" — answers vector search alone can\'t reach. Benchmarked ' +
       'end-to-end on a 240-page rich-prose corpus: Recall@5 83% → 95%, Precision@5 ' +
       '39% → 45%, +30 more correct answers in the agent\'s top-5. Graph-only F1: ' +
-      '86.6% vs grep\'s 57.8% (+28.8 pts). See docs/benchmarks/2026-04-18-brainbench-v1.md.',
+      '86.6% vs grep\'s 57.8% (+28.8 pts). See github.com/garrytan/gbrain-evals.',
   },
   orchestrator,
 };
diff --git a/src/core/link-extraction.ts b/src/core/link-extraction.ts
index 921bbf5eb..ffb66d24e 100644
--- a/src/core/link-extraction.ts
+++ b/src/core/link-extraction.ts
@@ -353,7 +353,20 @@ function excerpt(s: string, idx: number, width: number): string {
 //     explicit "advisor"/"advise" rooting.
 
 // Employment context: position + at/of, or explicit work verbs.
-const WORKS_AT_RE = /\b(?:CEO of|CTO of|COO of|CFO of|CMO of|CRO of|VP at|VP of|VPs? Engineering|VPs? Product|works at|worked at|working at|employed by|employed at|joined as|joined the team|engineer at|engineer for|director at|director of|head of|leads engineering|leads product|currently at|previously at|previously worked at|spent .* (?:years|months) at|stint at|tenure at)\b/i;
+//
+// v0.10.5 additions (drive works_at 58% → >85% on rich prose):
+//   - Role-prefixed engineer patterns: "senior engineer at", "staff engineer at",
+//     "principal engineer at", "lead engineer at". Current "engineer at" only
+//     hits if the word "engineer" is immediately adjacent; prose often uses
+//     rank-qualified forms.
+//   - Generic role patterns: "backend engineer at", "frontend engineer at",
+//     "ML engineer at", "data engineer at", "full-stack engineer at".
+//   - Broader role verbs: "manages engineering at", "running product at",
+//     "leads the [team] at", "heads up engineering at".
+//   - Possessive time: "his time at", "her time at", "their time at", "my time at".
+//   - Role noun forms: "role at", "tenure as", "stint as", "position at".
+//   - Promoted/staff-engineer forms: "promoted to (staff|senior|principal) engineer at".
+const WORKS_AT_RE = /\b(?:CEO of|CTO of|COO of|CFO of|CMO of|CRO of|VP at|VP of|VPs? Engineering|VPs? Product|works at|worked at|working at|employed by|employed at|joined as|joined the team|engineer at|engineer for|director at|director of|head of|heads up .{0,20} at|leads engineering|leads product|leads the .{0,20} (?:team|org) at|manages engineering at|manages product at|running (?:engineering|product|design) at|currently at|previously at|previously worked at|spent .* (?:years|months) at|stint at|stint as|tenure at|tenure as|role at|position at|(?:senior|staff|principal|lead|backend|frontend|full-?stack|ML|data|security) engineer at|promoted to (?:senior|staff|principal|lead) .{0,20} at|(?:his|her|their|my) time at)\b/i;
 
 // Investment context. Order patterns from most-specific to least to keep
 // regex efficient. Includes funding-round verbs ("led the seed", "led X's
@@ -370,13 +383,39 @@ const FOUNDED_RE = /\b(?:founded|co-?founded|started the company|incorporated|fo
 // Advise context: must be rooted in "advisor"/"advise" (investors also sit on
 // boards). Keep "board advisor" / "advisory board" but drop generic "board
 // member" / "sits on the board" which over-matches.
-const ADVISES_RE = /\b(?:advises|advised|advisor (?:to|at|for|of)|advisory (?:board|role|position)|board advisor|on .{0,20} advisory board|joined .{0,20} advisory board)\b/i;
+//
+// v0.10.5 additions (drive advises 41% → >85% on rich prose):
+//   - Advisory capacity phrasings: "in an advisory capacity", "advisory engagement",
+//     "advisory partnership", "advisory contract", "advisory relationship".
+//   - "as an advisor" form: joined/serves/brought on "as an advisor" / "as a
+//     security advisor" / "as a technical advisor" / "as an industry advisor".
+//   - "consults for / consulting role": advisor-adjacent verbs that appear in
+//     narratives where the direct "advises" verb isn't used.
+//   - Advisor-qualified: "strategic advisor to|at", "technical advisor to|at",
+//     "security advisor to|at", "product advisor to|at", "industry advisor".
+const ADVISES_RE = /\b(?:advises|advised|advisor (?:to|at|for|of)|advisory (?:board|role|position|capacity|engagement|partnership|contract|relationship|work)|board advisor|on .{0,20} advisory board|joined .{0,20} advisory board|in an? advisory (?:capacity|role|position)|as an? (?:advisor|security advisor|technical advisor|strategic advisor|industry advisor|product advisor|board advisor|senior advisor)|(?:strategic|technical|security|product|industry|senior|board) advisor (?:to|at|for|of)|consults for|consulting role (?:at|with))\b/i;
 
 // Page-role detection: if the source page describes a partner/investor at
 // page level, that's a strong prior for outbound company refs being
 // invested_in even when per-edge context lacks explicit investment verbs.
 const PARTNER_ROLE_RE = /\b(?:partner at|partner of|venture partner|VC partner|invested early|investor at|investor in|portfolio|venture capital|early-stage investor|seed investor|fund [A-Z]|invests across|backs companies)\b/i;
-const ADVISOR_ROLE_RE = /\b(?:full-time advisor|professional advisor|advises (?:multiple|several|various))\b/i;
+
+// Advisor role prior: fires when the page-level description indicates the
+// person IS an advisor (not just mentions advising). Broadened in v0.10.5
+// from "full-time/professional/advises multiple" to catch any page that
+// self-identifies the subject as an advisor.
+const ADVISOR_ROLE_RE = /\b(?:full-time advisor|professional advisor|advises (?:multiple|several|various)|is an? (?:advisor|security advisor|technical advisor|strategic advisor|industry advisor|product advisor|senior advisor)|took on advisory roles|(?:her|his|their) advisory (?:work|role|engagement|portfolio)|serves as (?:an )?advisor)\b/i;
+
+// Employee role prior (new in v0.10.5): fires when the page-level description
+// indicates the person IS an employee (senior/staff/lead engineer, director,
+// head, etc.) at some company. Biases outbound company refs on that page
+// toward works_at when per-edge verbs are absent (e.g. possessive phrasings
+// "her work on Delta's pipeline..." where the verb "works" doesn't appear
+// near the slug).
+//
+// Scope: only fires for person-page → company-page links. Companies' own
+// pages mentioning their employees use the page-role layer differently.
+const EMPLOYEE_ROLE_RE = /\b(?:is an? (?:senior|staff|principal|lead|backend|frontend|full-?stack|ML|data|security|DevOps|platform)? ?engineer at|is an? (?:senior|staff|principal|lead)? ?(?:developer|designer|product manager|engineering manager|director|VP) (?:at|of)|holds? the (?:CTO|CEO|CFO|COO|CMO|CRO|VP) (?:role|position|seat|title) at|is the (?:CTO|CEO|CFO|COO|CMO|CRO) of|employee at|on the team at|works on .{0,30} at)\b/i;
 
 /**
  * Infer link_type from page context. Deterministic regex heuristics, no LLM.
@@ -408,9 +447,15 @@ export function inferLinkType(pageType: PageType, context: string, globalContext
   // about VC topics naturally contain "venture capital" in their text, but
   // their company refs are mentions, not investments. Partner pages mentioning
   // other people (co-investors, friends) should also stay as mentions.
+  //
+  // Precedence within priors: investor > advisor > employee. Investors often
+  // also sit on boards ("board seat at portfolio company") which a naive
+  // employee/advisor match would mis-classify; keep investor first so those
+  // phrasings resolve correctly.
   if (pageType === 'person' && globalContext && targetSlug?.startsWith('companies/')) {
     if (PARTNER_ROLE_RE.test(globalContext)) return 'invested_in';
     if (ADVISOR_ROLE_RE.test(globalContext)) return 'advises';
+    if (EMPLOYEE_ROLE_RE.test(globalContext)) return 'works_at';
   }
   return 'mentions';
 }
diff --git a/src/core/markdown.ts b/src/core/markdown.ts
index 0b7f13b68..11752ed6c 100644
--- a/src/core/markdown.ts
+++ b/src/core/markdown.ts
@@ -169,6 +169,13 @@ function inferType(filePath?: string): PageType {
   if (lower.includes('/projects/') || lower.includes('/project/')) return 'project';
   if (lower.includes('/sources/') || lower.includes('/source/')) return 'source';
   if (lower.includes('/media/')) return 'media';
+  // BrainBench v1 amara-life-v1 corpus directories. One-slash slug convention
+  // means source paths look like `emails/em-0001.md`, `slack/sl-0037.md`, etc.
+  if (lower.includes('/emails/') || lower.includes('/email/')) return 'email';
+  if (lower.includes('/slack/')) return 'slack';
+  if (lower.includes('/cal/') || lower.includes('/calendar/')) return 'calendar-event';
+  if (lower.includes('/notes/') || lower.includes('/note/')) return 'note';
+  if (lower.includes('/meetings/') || lower.includes('/meeting/')) return 'meeting';
   return 'concept';
 }
 
diff --git a/src/core/types.ts b/src/core/types.ts
index f91839ab1..f45a79254 100644
--- a/src/core/types.ts
+++ b/src/core/types.ts
@@ -1,5 +1,9 @@
 // Page types
-export type PageType = 'person' | 'company' | 'deal' | 'yc' | 'civic' | 'project' | 'concept' | 'source' | 'media' | 'writing' | 'analysis' | 'guide' | 'hardware' | 'architecture' | 'meeting' | 'note';
+// email | slack | calendar-event: native Page types for inbox/chat/calendar
+// ingest (and the amara-life-v1 eval corpus in the sibling gbrain-evals repo).
+// Previously these collapsed into `source`, which lost workflow semantics
+// (e.g. "attended meetings" vs "received emails").
+export type PageType = 'person' | 'company' | 'deal' | 'yc' | 'civic' | 'project' | 'concept' | 'source' | 'media' | 'writing' | 'analysis' | 'guide' | 'hardware' | 'architecture' | 'meeting' | 'note' | 'email' | 'slack' | 'calendar-event';
 
 export interface Page {
   id: number;
diff --git a/test/brain-allowlist.test.ts b/test/brain-allowlist.test.ts
index 6237c4de5..2cd272bf8 100644
--- a/test/brain-allowlist.test.ts
+++ b/test/brain-allowlist.test.ts
@@ -27,11 +27,11 @@ beforeAll(async () => {
   engine = new PGLiteEngine();
   await engine.connect({ database_url: '' });
   await engine.initSchema();
-});
+}, 60_000);
 
 afterAll(async () => {
   await engine.disconnect();
-});
+}, 60_000);
 
 beforeEach(async () => {
   await engine.executeRaw('DELETE FROM pages');
diff --git a/test/core/cycle.test.ts b/test/core/cycle.test.ts
index 4f96fe18d..308a0325c 100644
--- a/test/core/cycle.test.ts
+++ b/test/core/cycle.test.ts
@@ -135,11 +135,11 @@ beforeAll(async () => {
   sharedEngine = new PGLiteEngine();
   await sharedEngine.connect({});
   await sharedEngine.initSchema();
-});
+}, 60_000);
 
 afterAll(async () => {
   await sharedEngine.disconnect();
-});
+}, 60_000);
 
 beforeEach(() => {
   lintCalls = [];
diff --git a/test/dream.test.ts b/test/dream.test.ts
index 768577aeb..01127d1a3 100644
--- a/test/dream.test.ts
+++ b/test/dream.test.ts
@@ -55,12 +55,12 @@ describe('runDream — brainDir resolution', () => {
   beforeEach(async () => {
     repo = makeGitRepo();
     engine = await makePGLite();
-  });
+  }, 60_000);
 
   afterEach(async () => {
     await engine.disconnect();
     rmSync(repo, { recursive: true, force: true });
-  });
+  }, 60_000);
 
   test('explicit --dir takes precedence over engine config', async () => {
     await engine.setConfig('sync.repo_path', '/configured/dir');
@@ -112,12 +112,12 @@ describe('runDream — --phase <name> restricts the cycle', () => {
   beforeEach(async () => {
     repo = makeGitRepo();
     engine = await makePGLite();
-  });
+  }, 60_000);
 
   afterEach(async () => {
     await engine.disconnect();
     rmSync(repo, { recursive: true, force: true });
-  });
+  }, 60_000);
 
   test('--phase lint produces a report with exactly one phase = lint', async () => {
     const report = await runDream(engine, ['--dir', repo, '--phase', 'lint', '--json']);
@@ -160,12 +160,12 @@ describe('runDream — output format', () => {
   beforeEach(async () => {
     repo = makeGitRepo();
     engine = await makePGLite();
-  });
+  }, 60_000);
 
   afterEach(async () => {
     await engine.disconnect();
     rmSync(repo, { recursive: true, force: true });
-  });
+  }, 60_000);
 
   test('--json emits parsable CycleReport JSON with schema_version', async () => {
     const lines: string[] = [];
@@ -198,12 +198,12 @@ describe('runDream — dry-run propagates through to runCycle', () => {
   beforeEach(async () => {
     repo = makeGitRepo();
     engine = await makePGLite();
-  });
+  }, 60_000);
 
   afterEach(async () => {
     await engine.disconnect();
     rmSync(repo, { recursive: true, force: true });
-  });
+  }, 60_000);
 
   test('--dry-run produces a report where no DB-mutating work happened', async () => {
     // Before: empty pages table.
@@ -227,12 +227,12 @@ describe('runDream — exit-code semantics', () => {
   beforeEach(async () => {
     repo = makeGitRepo();
     engine = await makePGLite();
-  });
+  }, 60_000);
 
   afterEach(async () => {
     await engine.disconnect();
     rmSync(repo, { recursive: true, force: true });
-  });
+  }, 60_000);
 
   test('clean/ok/partial statuses do not call process.exit', async () => {
     const spy = spyOn(process, 'exit').mockImplementation(() => { throw new Error('UNEXPECTED_EXIT'); });
diff --git a/test/extract-db.test.ts b/test/extract-db.test.ts
index 50364e7a6..ec860a7c5 100644
--- a/test/extract-db.test.ts
+++ b/test/extract-db.test.ts
@@ -19,11 +19,11 @@ beforeAll(async () => {
   engine = new PGLiteEngine();
   await engine.connect({});
   await engine.initSchema();
-});
+}, 60_000);
 
 afterAll(async () => {
   await engine.disconnect();
-});
+}, 60_000);
 
 async function truncateAll() {
   for (const t of ['content_chunks', 'links', 'tags', 'raw_data', 'timeline_entries', 'page_versions', 'ingest_log', 'pages']) {
diff --git a/test/link-extraction.test.ts b/test/link-extraction.test.ts
index 5f18dd71c..888292991 100644
--- a/test/link-extraction.test.ts
+++ b/test/link-extraction.test.ts
@@ -209,6 +209,123 @@ describe('inferLinkType', () => {
   test('media page -> mentions (not attended)', () => {
     expect(inferLinkType('media', 'Alice attended the workshop.')).toBe('mentions');
   });
+
+  // ─── v0.10.5: works_at residuals (drive 58% → >85% on rich prose) ───
+
+  test('v0.10.5 works_at: rank-prefixed engineer at', () => {
+    expect(inferLinkType('person', 'Adam is a senior engineer at Delta.')).toBe('works_at');
+    expect(inferLinkType('person', 'She is a staff engineer at Stripe.')).toBe('works_at');
+    expect(inferLinkType('person', 'Promoted to principal engineer at Acme.')).toBe('works_at');
+  });
+
+  test('v0.10.5 works_at: discipline-prefixed engineer at', () => {
+    expect(inferLinkType('person', 'Backend engineer at NovaPay.')).toBe('works_at');
+    expect(inferLinkType('person', 'Full-stack engineer at Vox.')).toBe('works_at');
+    expect(inferLinkType('person', 'ML engineer at DeepMind.')).toBe('works_at');
+    expect(inferLinkType('person', 'Security engineer at Stripe.')).toBe('works_at');
+  });
+
+  test('v0.10.5 works_at: possessive time at', () => {
+    expect(inferLinkType('person', 'During her time at Goldman, she built the team.')).toBe('works_at');
+    expect(inferLinkType('person', 'His time at Delta taught him systems thinking.')).toBe('works_at');
+  });
+
+  test('v0.10.5 works_at: leadership verbs beyond "leads engineering"', () => {
+    expect(inferLinkType('person', 'She heads up design at Beta.')).toBe('works_at');
+    expect(inferLinkType('person', 'He manages engineering at Gamma.')).toBe('works_at');
+    expect(inferLinkType('person', 'She leads the platform team at Delta.')).toBe('works_at');
+    expect(inferLinkType('person', 'Running product at Stripe.')).toBe('works_at');
+  });
+
+  test('v0.10.5 works_at: tenure/stint/role as', () => {
+    expect(inferLinkType('person', 'Her tenure as head of engineering was short.')).toBe('works_at');
+    expect(inferLinkType('person', 'A brief stint as VP of Product.')).toBe('works_at');
+    expect(inferLinkType('person', 'His role at Delta was to unblock the pipeline team.')).toBe('works_at');
+  });
+
+  test('v0.10.5 works_at: page-role employee prior for ambiguous context', () => {
+    // Per-edge context doesn't mention a work verb, but globalContext establishes
+    // the person IS a senior engineer at a company. The employee role prior
+    // should bias outbound company refs toward works_at.
+    const globalContext = 'Adam Lopez is a senior engineer at Delta. His work is excellent.';
+    const perEdgeContext = 'Adam is excellent.';  // no work verb in the window
+    expect(inferLinkType('person', perEdgeContext, globalContext, 'companies/delta-3')).toBe('works_at');
+  });
+
+  test('v0.10.5 works_at: page-role CTO-of prior', () => {
+    const globalContext = 'Beth is the CTO of Prism, shipping their platform.';
+    const perEdgeContext = 'Beth is shipping.';  // no work verb near slug
+    expect(inferLinkType('person', perEdgeContext, globalContext, 'companies/prism-43')).toBe('works_at');
+  });
+
+  // ─── v0.10.5: advises residuals (drive 41% → >85% on rich prose) ───
+
+  test('v0.10.5 advises: "as an advisor" / "as a security advisor"', () => {
+    expect(inferLinkType('person', 'Joined Acme as an advisor in 2022.')).toBe('advises');
+    expect(inferLinkType('person', 'Brought on as a security advisor.')).toBe('advises');
+    expect(inferLinkType('person', 'Serves as a technical advisor to the team.')).toBe('advises');
+  });
+
+  test('v0.10.5 advises: prefixed advisor (security advisor to X)', () => {
+    expect(inferLinkType('person', 'She is the security advisor to Orbit Labs.')).toBe('advises');
+    expect(inferLinkType('person', 'He is a strategic advisor at Prism.')).toBe('advises');
+    expect(inferLinkType('person', 'Product advisor to several early-stage startups.')).toBe('advises');
+  });
+
+  test('v0.10.5 advises: "in an advisory capacity"', () => {
+    expect(inferLinkType('person', 'Engaged with Prism in an advisory capacity.')).toBe('advises');
+    expect(inferLinkType('person', 'Continued in an advisory role through 2024.')).toBe('advises');
+  });
+
+  test('v0.10.5 advises: advisory engagement / partnership / contract', () => {
+    expect(inferLinkType('person', 'Began a formal advisory engagement with Prism.')).toBe('advises');
+    expect(inferLinkType('person', 'Signed an advisory contract last year.')).toBe('advises');
+    expect(inferLinkType('person', 'Multi-year advisory partnership with Beta.')).toBe('advises');
+  });
+
+  test('v0.10.5 advises: page-role "is an advisor" prior', () => {
+    // Per-edge window has no advisor verb (just possessive "her work"), but
+    // page-level establishes the subject IS an advisor. Prior should fire.
+    const globalContext = 'Alice Davis is an advisor at Prism. Her work has been invaluable.';
+    const perEdgeContext = 'Alice Davis has been invaluable.';  // no advise verb in window
+    expect(inferLinkType('person', perEdgeContext, globalContext, 'companies/prism-43')).toBe('advises');
+  });
+
+  test('v0.10.5 advises: "serves as advisor" page prior', () => {
+    // Avoid "portfolio" in global context since that trips PARTNER_ROLE_RE.
+    // Real advisor pages rarely use "portfolio" (that's a partner word).
+    const globalContext = 'Beth serves as advisor to three early-stage startups.';
+    const perEdgeContext = 'Beth sees Acme regularly.';
+    expect(inferLinkType('person', perEdgeContext, globalContext, 'companies/acme')).toBe('advises');
+  });
+
+  // ─── Regression guards: v0.10.5 expansions must not break tightened rules ───
+
+  test('v0.10.5 regression: generic "board member" still resolves to mentions', () => {
+    // This was the v0.10.4 tightening. The expanded ADVISES_RE must not
+    // re-introduce the false-positive on partner bios.
+    expect(inferLinkType('person', 'Jane is a board member at Beta Health.')).toBe('mentions');
+  });
+
+  test('v0.10.5 regression: "sits on the board" still mentions (not advises)', () => {
+    expect(inferLinkType('person', 'She sits on the board of Acme.')).toBe('mentions');
+  });
+
+  test('v0.10.5 regression: "backs companies" still resolves to invested_in via partner prior', () => {
+    // Partner prior takes precedence over employee prior.
+    const globalContext = 'Wendy is a venture partner who backs companies at the seed stage. Her portfolio is diverse.';
+    const perEdgeContext = 'Wendy recently discussed Cipher.';
+    expect(inferLinkType('person', perEdgeContext, globalContext, 'companies/cipher-13')).toBe('invested_in');
+  });
+
+  test('v0.10.5 regression: partner + advisor co-mention stays invested_in for investee', () => {
+    // If someone is both a partner AND mentions advisory work, the outbound
+    // companies should lean toward invested_in (partner precedence). This
+    // protects against a common pattern where partners say "I also advise X".
+    const globalContext = 'Jane is a partner at Accel. She also advises multiple startups.';
+    const perEdgeContext = 'Jane has worked with Acme.';
+    expect(inferLinkType('person', perEdgeContext, globalContext, 'companies/acme')).toBe('invested_in');
+  });
 });
 
 // ─── parseTimelineEntries ──────────────────────────────────────
diff --git a/test/markdown.test.ts b/test/markdown.test.ts
index 52c468888..2d6f165ba 100644
--- a/test/markdown.test.ts
+++ b/test/markdown.test.ts
@@ -87,6 +87,26 @@ Content
     const parsed = parseMarkdown(md, 'concepts/do-things-that-dont-scale.md');
     expect(parsed.slug).toBe('concepts/do-things-that-dont-scale');
   });
+
+  // v0.20: BrainBench / native inbox-chat-calendar Page types. These 5 directory
+  // heuristics exercise PageType 'email | slack | calendar-event | note | meeting'
+  // which were added for amara-life-v1 ingest but are useful for any gbrain user
+  // ingesting an inbox dump, Slack export, iCal, meeting transcript, or daily notes.
+  test.each([
+    ['emails/em-0001.md', 'email'],
+    ['email/em-0001.md', 'email'],
+    ['slack/sl-0037.md', 'slack'],
+    ['cal/evt-0042.md', 'calendar-event'],
+    ['calendar/evt-0042.md', 'calendar-event'],
+    ['notes/2026-04-standup.md', 'note'],
+    ['note/2026-04-standup.md', 'note'],
+    ['meetings/mtg-0003.md', 'meeting'],
+    ['meeting/mtg-0003.md', 'meeting'],
+  ] as const)('infers type %s -> %s', (path, expectedType) => {
+    const md = `---\ntitle: Fixture\n---\nBody\n`;
+    const parsed = parseMarkdown(md, path);
+    expect(parsed.type).toBe(expectedType);
+  });
 });
 
 describe('splitBody', () => {
diff --git a/test/multi-source-integration.test.ts b/test/multi-source-integration.test.ts
index f6fb9ba18..b5c301d76 100644
--- a/test/multi-source-integration.test.ts
+++ b/test/multi-source-integration.test.ts
@@ -27,11 +27,11 @@ beforeAll(async () => {
   engine = new PGLiteEngine();
   await engine.connect({ type: 'pglite' } as never);
   await engine.initSchema();
-});
+}, 60_000);
 
 afterAll(async () => {
   await engine.disconnect();
-});
+}, 60_000);
 
 describe('v0.18.0 — sources table seeded with default row on fresh PGLite', () => {
   test("sources('default') exists after initSchema + migration", async () => {
diff --git a/test/orphans.test.ts b/test/orphans.test.ts
index d2bd6af89..762a5feb7 100644
--- a/test/orphans.test.ts
+++ b/test/orphans.test.ts
@@ -216,11 +216,11 @@ describe('findOrphans (engine-injected)', () => {
     engine = new PGLiteEngine();
     await engine.connect({});
     await engine.initSchema();
-  });
+  }, 60_000);
 
   afterEach(async () => {
     await engine.disconnect();
-  });
+  }, 60_000);
 
   test('returns pages with no inbound links, excluding pseudo-pages', async () => {
     // Build a tiny brain: alice links to bob. alice is an orphan (nothing