mirror of
https://github.com/garrytan/gbrain.git
synced 2026-07-28 14:59:47 +00:00
* feat(bench): add baseline-file, qrels-file, correctness-gate shared modules
v0.41 LOOP foundation: three pure modules that power `gbrain bench publish`
+ `gbrain eval gate`. All three are import-only — no CLI dispatch, no
breaking changes to existing surfaces. Tested in isolation (34 cases).
- src/core/bench/baseline-file.ts (~190 LOC): single source of truth for
the .baseline.ndjson file shape. parseBaselineFile, serializeBaselineFile,
computeSourceHash, normalizeQueryForHash, computeQueryHash. Body rows
stamped with schema_version: 1 so existing eval-replay parser accepts
them unchanged.
- src/core/bench/qrels-file.ts (~210 LOC): pure parser + math for the
.qrels.json shape. Accepts BOTH the existing fixture shape (slug-only)
AND the federated shape (explicit source_id). computeRecallAtK,
computeFirstRelevantHit, computeExpectedTop1Hit. Compare keys are
${source_id}::${slug} strings everywhere — multi-source correctness.
- src/core/bench/correctness-gate.ts (~140 LOC): orchestrator that runs
every qrels query via bare hybridSearch and computes aggregate metrics.
Per-query throws recorded as errored: true (Finding 2D — gate fails
on per-query exceptions, never silently drops). Injectable searchFn
test seam.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(eval-replay): skip baseline_metadata header + expose replayCore
Two surgical changes to existing eval-replay so `gbrain eval gate` can
call replay in-process without spawning a subprocess (which would run
the INSTALLED gbrain, not the workspace version — codex round-2 #7
caught this drift risk on source-tree CI runs).
- parseNdjson now skips lines where _kind === 'baseline_metadata'.
Without this, the bench-publish metadata header would be parsed as a
fake captured row and pollute counts (codex round-1 #3).
- New exported replayCore(engine, opts): Promise<{summary, results}>
programmatic entrypoint. Existing CLI runEvalReplay now wraps it.
ReplaySummary interface also exported for eval-gate consumers.
IRON-RULE regression pinned by test/eval-replay-metadata-skip.test.ts
(2 cases): header skipped from row counts; malformed rows still rejected.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(bench): add `gbrain bench publish` CLI verb
The LOOP-closing verb. Turns captured eval rows (gbrain eval export) into
a baseline file (.baseline.ndjson) consumed by gbrain eval gate --baseline.
Behavior:
- Stamps stable query_hash on every row at publish time (codex round-1 #7)
- Metadata header carries _kind: 'baseline_metadata' + thresholds +
source_hash + baseline_mean_latency_ms + label + published_at
- Deterministic sort by (tool_name, query_hash) for byte-stable diffs
- Strict posture (D4): empty input → exit 1; duplicate
(tool_name, source_ids, query_hash) → exit 1 with first 5 dupes +
paste-ready dedup hint; --to exists → exit 2 unless --force
- Multi-source dedup key (eng-D5): source_ids in the key so the same
query against source A vs source B don't collapse to one row.
Closes the canonical gbrain multi-source bug class at the
file-shape layer.
- Audit JSONL at ~/.gbrain/audit/bench-publish-YYYY-Www.jsonl via
shared audit-writer primitive.
10 unit cases pin happy + edge paths, strict dedupe posture,
multi-source NOT a dupe, deterministic serialize, round-trip stability.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(eval): add `gbrain eval gate` two-gate CI verb
The CI-gating verb. Two gating paths (CEO D8 + eng D6/D7):
- Regression gate (--baseline X.baseline.ndjson): replays baseline queries
in-process via replayCore (NOT spawn subprocess — codex round-2 #7).
Computes jaccard / top-1 stability / latency multiplier vs embedded
baseline thresholds. Catches retrieval REGRESSIONS during refactors.
- Correctness gate (--qrels Y.qrels.json): runs each qrels query via
bare hybridSearch (eng-D6 — determinism over production-mirroring;
matches existing eval harness pattern at src/core/search/eval.ts:242).
Computes recall@K + first_relevant_hit_rate + expected_top1_hit_rate.
Catches retrieval QUALITY drops against known-right answers.
Both can be passed together; both must pass for verdict 'pass'. At least
one required (usage error otherwise).
Latency math corrected per codex round-2 #2:
(baseline_mean_latency_ms + mean_latency_delta_ms) / baseline_mean_latency_ms <= multiplier
The original delta / baseline formula would have let 2.5x slowdowns pass
at multiplier=2.0.
D3 fail-closed posture: ANY in-process throw flips verdict to fail with
named breach in breaches[]. Never silently exits 0.
Exit codes: 0 PASS, 1 FAIL (regression OR throw), 2 USAGE.
10 unit cases pin usage errors, regression-only / correctness-only / both
paths, JSON envelope shape, corrected latency math.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(autopilot): wire nightly quality probe (opt-in, off by default)
Closes the v0.40.1.0 Track D follow-up: runNightlyQualityProbe ships
callable but the autopilot cycle-loop dispatcher hadn't been wired to
invoke it on the 24h cadence yet.
- src/commands/autopilot.ts (tick body): invokes runNightlyQualityProbe
when cfg.autopilot.nightly_quality_probe.enabled === true.
Per eng-D10 (codex round-1 #11): NO scheduler-side rate-limit check.
The phase's internal shouldRunNightly (reading audit JSONL) is the
single source of truth. Probe call wrapped in try/catch that logs to
stderr and DOES NOT bump consecutiveErrors (probe failure is
informational, never crashes the loop).
- src/core/cycle/nightly-probe-adapters.ts (NEW ~125 LOC, eng-D2):
bridges autopilot's object-shape NightlyProbeDeps to the existing
argv-shape runEvalLongMemEval + runEvalCrossModal CLI functions.
Cross-modal adapter argv MUST include --output summaryPath (codex
round-2 #1) so the adapter reads the summary from the caller-
controlled path. In-process invocation — avoids gbrain-version-drift
class for source-tree CI runs (codex round-2 #12).
- src/core/config.ts: added autopilot.nightly_quality_probe to
GBrainConfig interface (typecheck gate).
Default OFF — opt-in via:
gbrain config set autopilot.nightly_quality_probe.enabled true
Cost cap default $5/run × 30 nights ≈ $150/month worst-case per brain.
Expected real cost ~$0.35/night × 30 ≈ $10.50/month.
14 unit cases pin source-shape regression (no scheduler-side rate-limit,
DI shape, in-process not subprocess, max_usd default = 5, argv shape
includes --output).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(e2e): full capture → publish → gate LOOP integration (PGLite)
Hermetic end-to-end test of the v0.41 LOOP per eng-D5. Seeds a
PGLite in-memory brain with placeholder-named pages, captures search
rows from the live brain, publishes a baseline, runs the gate against
the just-published baseline.
4 cases:
- self-gate against just-published baseline returns PASS (LOOP closes)
- perturbed retrieved_slugs → jaccard drops → exit 1 with named breach
- malformed baseline → exit 1 fail-closed (D3 IRON-RULE — pre-D3 bug
would have silently exited 0)
- byte-stable round-trip: serialize → parse → re-serialize identical
Uses tool_name='search' (bare keyword) for captured rows so replay
runs hermetically without embedding-provider dependencies.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(eval-longmemeval): bump warm-create p50 gate 1500ms → 2500ms
CI runner observed p50 above 1500ms under parallel test load (8-way
shard × PGLite WASM contention). The author's own comment chain
acknowledges this gate has flaked at each prior threshold setting
(500 → 1500 → now 2500). 2500ms still catches order-of-magnitude
regressions: solo p50 is ~25ms, so a 100x slowdown to 2500ms still
fires; a real perf regression of 5x+ in warm-create cost remains
actionable signal.
Caught by CI test shard 2 on PR #1352 (v0.41.0.0). Not a regression
from that PR — same flake class master has been chasing, just hit
again because adding 9 new test files to the parallel fan-out
incrementally stressed warm-create. Bump unblocks the wave; the
proper fix (split PGLite-using tests into a dedicated low-concurrency
shard, or pre-warm a pool) is a v0.42+ test-infra task.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* chore: bump version 0.41.0.0 → 0.41.1.0
Per /ship queue convention — this wave releases as a MINOR bump
(2nd digit) reflecting that the eval-loop wave adds new capability
surfaces (gbrain bench publish, gbrain eval gate, autopilot nightly
probe wiring) on top of v0.41's already-shipped feature set.
VERSION + package.json + CHANGELOG header + "To take advantage" line
all updated together. Trio agrees on 0.41.1.0.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
599 lines
23 KiB
Markdown
599 lines
23 KiB
Markdown
# Running real-world eval benchmarks against your gbrain changes
|
||
|
||
Audience: gbrain maintainers and contributors. If you're touching retrieval
|
||
(search, ranking, embeddings, intent classification, query expansion, source
|
||
boost, hybrid fusion), this is the doc.
|
||
|
||
For the **NDJSON wire format** consumed by gbrain-evals, see
|
||
[`eval-capture.md`](./eval-capture.md). This doc is the human dev loop
|
||
that lives on top of that format.
|
||
|
||
## v0.41 update — the LOOP is now real
|
||
|
||
Before v0.41, you could capture eval rows and replay them but nothing
|
||
stitched them into a gate. `gbrain bench publish` + `gbrain eval gate`
|
||
close the loop. Two gates:
|
||
|
||
- **Regression gate** (`--baseline X.baseline.ndjson`): replays a baseline
|
||
you captured against your current brain. Catches: "did my refactor break
|
||
search?" Compares jaccard / top-1 stability / latency multiplier.
|
||
- **Correctness gate** (`--qrels Y.qrels.json`): runs known-right queries
|
||
against your current brain via bare `hybridSearch`. Catches: "is my
|
||
retrieval actually any good?" Computes recall@K, first-relevant-hit-rate,
|
||
expected_top1-hit-rate.
|
||
|
||
Both can be passed together; both must pass for verdict `pass`. At least
|
||
one is required.
|
||
|
||
### The full LOOP for your own brain
|
||
|
||
```bash
|
||
# 1. Capture (one-time; uses queries already in eval_candidates)
|
||
gbrain eval export --limit 200 --tool query > /tmp/captured.ndjson
|
||
|
||
# 2. Publish a baseline
|
||
mkdir -p ~/.gbrain/baselines
|
||
gbrain bench publish --from /tmp/captured.ndjson --to ~/.gbrain/baselines/personal.baseline.ndjson --label "personal-$(date +%Y%m%d)"
|
||
|
||
# 3. Gate against it
|
||
gbrain eval gate --baseline ~/.gbrain/baselines/personal.baseline.ndjson
|
||
```
|
||
|
||
### Privacy posture (D9)
|
||
|
||
**Public baselines in `gbrain-evals` are hermetic-synthetic ONLY.** Real
|
||
user captures stay local in `~/.gbrain/baselines/`. The boundary is
|
||
enforced at the file source, not by post-hoc scrubbing. If you publish a
|
||
baseline to `gbrain-evals`, generate it from a fixture-seeded test brain
|
||
(placeholder names like `alice-example`, `widget-co-example`) — never
|
||
from a real user's `eval_candidates` table.
|
||
|
||
### Deterministic-pipeline disclosure
|
||
|
||
`gbrain eval gate --qrels` uses bare `hybridSearch` (not the production
|
||
`query` op handler). This is deliberate: gates need to be deterministic in
|
||
CI. Production retrieval differs via the query cache, salience freshness,
|
||
expansion, etc. The gate measures retrieval quality with a fixed pipeline;
|
||
your users may see different results when the cache is warm.
|
||
|
||
### `.qrels.json` shape
|
||
|
||
Two equivalent representations per entry:
|
||
|
||
```json
|
||
{
|
||
"schema_version": 1,
|
||
"queries": [
|
||
{
|
||
"query_id": "q1",
|
||
"query": "fintech founder",
|
||
"relevant_slugs": ["people/alice-example"],
|
||
"first_relevant_slug": "people/alice-example"
|
||
}
|
||
]
|
||
}
|
||
```
|
||
|
||
For federated / multi-source brains, use the explicit shape (no defaults
|
||
to `source_id='default'`):
|
||
|
||
```json
|
||
{
|
||
"query_id": "q2",
|
||
"query": "anything",
|
||
"relevant": [
|
||
{"source_id": "host", "slug": "people/alice"},
|
||
{"source_id": "team-a", "slug": "people/alice"}
|
||
],
|
||
"expected_top1": {"source_id": "host", "slug": "people/alice"}
|
||
}
|
||
```
|
||
|
||
Without `source_id`, a hit from the wrong source could false-pass the
|
||
gate. The compare everywhere is `${source_id}::${slug}` strings.
|
||
|
||
### Example GitHub Actions workflow
|
||
|
||
```yaml
|
||
name: gbrain-eval-gate
|
||
on: [pull_request]
|
||
jobs:
|
||
gate:
|
||
runs-on: ubuntu-latest
|
||
steps:
|
||
- uses: actions/checkout@v4
|
||
- uses: oven-sh/setup-bun@v2
|
||
- run: bun install
|
||
- run: |
|
||
# Run both gates; CI fails on any breach.
|
||
gbrain eval gate \
|
||
--baseline gbrain-evals/baselines/v0.41-launch.baseline.ndjson \
|
||
--qrels gbrain-evals/qrels/v0.41-launch.qrels.json \
|
||
--json | tee /tmp/gate.json
|
||
```
|
||
|
||
---
|
||
|
||
## Prerequisite: turn on contributor mode
|
||
|
||
Capture is **off by default** for production users (privacy-positive — no
|
||
surprise data accumulation). Contributors flip it on with one line:
|
||
|
||
```bash
|
||
# In ~/.zshrc or ~/.bashrc:
|
||
export GBRAIN_CONTRIBUTOR_MODE=1
|
||
```
|
||
|
||
Verify:
|
||
|
||
```bash
|
||
gbrain query "anything" >/dev/null
|
||
psql $DATABASE_URL -c 'SELECT count(*) FROM eval_candidates' # should be > 0
|
||
```
|
||
|
||
To override (force on/off regardless of env var), edit `~/.gbrain/config.json`:
|
||
|
||
```json
|
||
{"eval": {"capture": true}} // force on
|
||
{"eval": {"capture": false}} // force off
|
||
```
|
||
|
||
Explicit config beats the env var both directions.
|
||
|
||
## The 4-command loop
|
||
|
||
```bash
|
||
# ① Capture: writes to eval_candidates whenever CONTRIBUTOR_MODE is set.
|
||
# Inspect what's been collected:
|
||
gbrain doctor # surfaces capture failures
|
||
psql $DATABASE_URL -c 'SELECT count(*) FROM eval_candidates'
|
||
|
||
# ② Snapshot: freeze a baseline before your code change.
|
||
gbrain eval export --since 7d > baseline.ndjson
|
||
|
||
# ③ Code change: do whatever you want — tune RRF_K, swap embed model, edit
|
||
# hybrid.ts, add a new boost source, change the intent classifier.
|
||
|
||
# ④ Replay: re-run every captured query against the current build.
|
||
gbrain eval replay --against baseline.ndjson
|
||
```
|
||
|
||
Output:
|
||
|
||
```
|
||
Replaying 247 captured queries…
|
||
...25/247
|
||
...50/247
|
||
...
|
||
Replayed 247 of 247 captured queries (0 skipped, 0 errored)
|
||
Mean Jaccard@k: 0.927
|
||
Top-1 stability: 91.5%
|
||
Mean latency Δ: +14ms (current vs captured)
|
||
|
||
Top 5 regression(s):
|
||
jaccard=0.20 captured=12 current=3 "find every reference to widget-co"
|
||
jaccard=0.43 captured=14 current=8 "show me everything tagged for review"
|
||
jaccard=0.50 captured=8 current=4 "what did alice say about the spec"
|
||
...
|
||
```
|
||
|
||
Three numbers tell you whether the change is safe to land:
|
||
|
||
| Metric | What it means | Healthy range |
|
||
|---|---|---|
|
||
| **Mean Jaccard@k** | Average overlap between captured retrieved slugs and current run's slugs. 1.0 = identical sets. | ≥0.85 for "neutral" changes. <0.7 means major retrieval shift. |
|
||
| **Top-1 stability** | Fraction of queries whose #1 result didn't change. | ≥85% for tuning passes. <70% means top-of-funnel broke. |
|
||
| **Mean latency Δ** | Current minus captured. Positive = slower now. | Within ±50ms of captured. >2× anywhere = regression alarm. |
|
||
|
||
## What it actually does
|
||
|
||
`gbrain eval replay` reads your NDJSON snapshot and, for each row:
|
||
|
||
1. Re-executes the same op (`searchKeyword` for `tool_name='search'`,
|
||
`hybridSearch` for `tool_name='query'`) with the captured `detail` and
|
||
`expand_enabled` values threaded back in.
|
||
2. Captures the current `retrieved_slugs` (deduped, in result order).
|
||
3. Computes set-Jaccard between captured and current slug sets.
|
||
4. Records top-1 match (was the #1 result the same slug?).
|
||
5. Records latency delta vs captured `latency_ms`.
|
||
|
||
It does NOT compute MRR or nDCG — those need ground-truth relevance labels,
|
||
not a baseline comparison. For metric-against-truth eval, use
|
||
`gbrain eval --qrels <path>` (the legacy IR-eval path, still supported). The
|
||
replay tool answers a different question: "did my code change move
|
||
retrieval, and which queries did it move most?"
|
||
|
||
For a third evaluation axis — public benchmark, ground-truth labels, full
|
||
question-answer pipeline (not just retrieval) — `gbrain eval longmemeval
|
||
<dataset.jsonl>` (v0.28.8) runs the LongMemEval benchmark against gbrain's
|
||
hybrid retrieval. Each question gets a clean in-memory PGLite, its haystack
|
||
imported, the question asked, the hypothesis emitted as JSONL — exactly the
|
||
shape LongMemEval's `evaluate_qa.py` consumes. Your `~/.gbrain` brain is
|
||
never opened. See `## Public benchmarks: LongMemEval` below.
|
||
|
||
## Best-effort by design
|
||
|
||
Replay is not pure. Three things can drift between capture and replay:
|
||
|
||
1. **Brain state** — your brain probably has more pages now than when the
|
||
snapshot was taken. Unless you explicitly seed a fixed corpus, mean
|
||
Jaccard will drop simply because new pages are eligible.
|
||
2. **Embedding source** — if you changed `OPENAI_API_KEY` between capture
|
||
and replay (or the embedding model rotated), vector-path results drift
|
||
even with identical code.
|
||
3. **Capture cap** — captured `retrieved_slugs` is a deduped set; it doesn't
|
||
preserve internal ranking metadata. Two tools can return the same slug
|
||
set with different scores — Jaccard will say 1.0, but a downstream
|
||
consumer that orders by score may behave differently.
|
||
|
||
The metrics are **regression alarms on real queries**, not a hash check.
|
||
Pair them with manual inspection of the top regressions.
|
||
|
||
## Cost
|
||
|
||
Every `query` row in the snapshot embeds the query string via OpenAI to run
|
||
the vector half of `hybridSearch`. Cost is identical to a normal `gbrain
|
||
query` invocation — text-embedding-3-large at OpenAI list price, batched
|
||
inside a single replay row.
|
||
|
||
If you're iterating locally and don't want to pay per change, use
|
||
`--limit 50` to cap rows replayed. The 50 most recent rows are usually
|
||
enough to catch direction; expand for the final pre-merge run.
|
||
|
||
```bash
|
||
# Iteration mode — 50 most recent queries
|
||
gbrain eval replay --against baseline.ndjson --limit 50
|
||
|
||
# Pre-merge — full snapshot
|
||
gbrain eval replay --against baseline.ndjson --top-regressions 20
|
||
```
|
||
|
||
## CI integration
|
||
|
||
```bash
|
||
gbrain eval replay --against baseline.ndjson --json > replay.json
|
||
jq -e '.summary.mean_jaccard >= 0.85' replay.json || exit 1
|
||
jq -e '.summary.top1_stability_rate >= 0.85' replay.json || exit 1
|
||
```
|
||
|
||
Stable JSON shape (schema_version: 1):
|
||
|
||
```json
|
||
{
|
||
"schema_version": 1,
|
||
"summary": {
|
||
"rows_total": 247,
|
||
"rows_replayed": 247,
|
||
"rows_skipped": 0,
|
||
"rows_errored": 0,
|
||
"mean_jaccard": 0.927,
|
||
"top1_stability_rate": 0.915,
|
||
"mean_latency_delta_ms": 14,
|
||
"rows_over_2x_latency": 0
|
||
}
|
||
}
|
||
```
|
||
|
||
`--verbose` adds a `results: [...]` array with one entry per replayed row
|
||
(useful for piping into jq or a notebook for deeper analysis).
|
||
|
||
## When to run this
|
||
|
||
Before merging anything that touches:
|
||
|
||
- `src/core/search/hybrid.ts` (RRF, fusion, dedup, two-pass retrieval)
|
||
- `src/core/search/source-boost.ts` / `sql-ranking.ts` (per-source ranking)
|
||
- `src/core/search/intent.ts` (auto-detail classification)
|
||
- `src/core/search/expansion.ts` (Haiku query expansion)
|
||
- `src/core/search/dedup.ts` (cross-page result collapse)
|
||
- `src/core/embedding.ts` or any embedding model swap
|
||
- `src/core/operations.ts` `query` or `search` op handlers (capture surface)
|
||
- `src/core/postgres-engine.ts` / `pglite-engine.ts` `searchKeyword` /
|
||
`searchVector` SQL
|
||
|
||
Skip for: schema-only migrations, doc changes, tests-only PRs, CLI ergonomics
|
||
that don't touch retrieval.
|
||
|
||
## Building your own corpus
|
||
|
||
If you don't have captured traffic yet (fresh install, can't dogfood for a
|
||
week before merging), you can hand-author an NDJSON file:
|
||
|
||
```jsonl
|
||
{"schema_version":1,"id":1,"tool_name":"query","query":"who is alice","retrieved_slugs":["people/alice","people/alice-bio"],"expand_enabled":false,"detail":null,"latency_ms":0,"remote":false}
|
||
{"schema_version":1,"id":2,"tool_name":"search","query":"acme deal","retrieved_slugs":["deals/acme-seed","companies/acme"],"latency_ms":0,"remote":false}
|
||
```
|
||
|
||
Then run `gbrain eval replay --against handcrafted.ndjson` to confirm the
|
||
authoritative slugs come back. This is the seam between the BrainBench-Real
|
||
pipeline (replay against live captures) and the BrainBench fixed-fixture
|
||
pipeline (`gbrain eval --qrels` with the sibling
|
||
[gbrain-evals](https://github.com/garrytan/gbrain-evals) corpus).
|
||
|
||
## Off-switch
|
||
|
||
Two ways to disable capture:
|
||
|
||
```bash
|
||
unset GBRAIN_CONTRIBUTOR_MODE # easy: just unset the env var
|
||
```
|
||
|
||
Or force off regardless of the env var via `~/.gbrain/config.json`:
|
||
|
||
```json
|
||
{"eval": {"capture": false}}
|
||
```
|
||
|
||
Existing `eval_candidates` rows stay until you `gbrain eval prune
|
||
--older-than 0d` (or just drop the table).
|
||
|
||
## Failure modes
|
||
|
||
| What you see | What it means |
|
||
|---|---|
|
||
| `Mean Jaccard@k: 0.4`, top regressions all in one source dir | Source boost or hard-exclude regression on that prefix |
|
||
| `Top-1 stability: 30%`, mean Jaccard still high | RRF tuning shifted the rank order without changing the set — re-tune `rrfK` |
|
||
| `Mean latency Δ: +500ms`, jaccard high | Vector path got slower; check embedding API or HNSW probes |
|
||
| `rows_errored > 0` | One or more queries threw. Inspect first 3 in human output, or `--json` to see all `error_message` fields |
|
||
| Many `skipped: empty query` | Capture ran on rows where someone passed empty `query` — check why those were captured |
|
||
|
||
## Public benchmarks: LongMemEval (v0.28.8)
|
||
|
||
`gbrain eval longmemeval` runs the public [LongMemEval](https://huggingface.co/datasets/xiaowu0162/longmemeval)
|
||
benchmark directly against gbrain's hybrid retrieval. Different evaluation
|
||
axis from `eval replay`: public dataset with ground-truth labels, end-to-end
|
||
question-answer pipeline, hermetic per-question brains.
|
||
|
||
```bash
|
||
# Download the dataset (visit the HF page in a browser; gated/manual download).
|
||
# Place longmemeval_oracle.json (or _s.json) somewhere local.
|
||
|
||
# Retrieval-only (no LLM answer-gen, fastest path, no Anthropic key needed):
|
||
gbrain eval longmemeval ./longmemeval_oracle.json --limit 50 --retrieval-only \
|
||
> /tmp/hypothesis.jsonl
|
||
|
||
# Full pipeline (Anthropic key required for answer-gen):
|
||
gbrain eval longmemeval ./longmemeval_oracle.json --limit 50 \
|
||
> /tmp/hypothesis.jsonl
|
||
|
||
# Score with LongMemEval's published evaluate_qa.py (not bundled — needs
|
||
# OpenAI gpt-4o per their spec):
|
||
python evaluate_qa.py /tmp/hypothesis.jsonl
|
||
```
|
||
|
||
### Architecture (read this if you're touching the harness)
|
||
|
||
- One in-memory PGLite per benchmark run via `createBenchmarkBrain` +
|
||
`withBenchmarkBrain`. Your `~/.gbrain` is never opened.
|
||
- Between questions: `TRUNCATE` over runtime-enumerated `pg_tables`, NOT a
|
||
hardcoded list — schema migrations don't silently leak data across
|
||
questions. Infrastructure tables (`sources`, `config`,
|
||
`gbrain_cycle_locks`, `subagent_rate_leases`) are preserved across resets.
|
||
- Sanitization parity: re-uses `INJECTION_PATTERNS` from
|
||
`src/core/think/sanitize.ts` so adding a new injection pattern
|
||
automatically covers takes AND benchmarks. One source of truth.
|
||
- Retrieved chat content is wrapped in `<chat_session id="..." date="...">`
|
||
framing; the answer-gen system prompt declares the content UNTRUSTED.
|
||
Same posture as `<take>` framing.
|
||
- LLM injection seam: `runEvalLongMemEval(args, {client?: ThinkLLMClient})`.
|
||
Tests stub the client so the full pipeline runs hermetically without any
|
||
API key.
|
||
|
||
### Flags
|
||
|
||
| Flag | Default | Purpose |
|
||
|---|---|---|
|
||
| `--limit N` | run all | Cap question count (iterate fast) |
|
||
| `--retrieval-only` | off | Emit retrieved chunks; no LLM answer-gen |
|
||
| `--keyword-only` | off | Disable vector path (debug retrieval issues) |
|
||
| `--expansion` | **off** | Multi-query expansion. Off by default for determinism (no per-query Haiku call). Pass to opt in. |
|
||
| `--top-k K` | 10 | Retrieval depth |
|
||
| `--model M` | resolved | Default resolves through `resolveModel()` 6-tier chain (`models.eval.longmemeval` config key) |
|
||
| `--output FILE` | stdout | Write hypothesis JSONL to file instead of stdout |
|
||
|
||
### Numbers
|
||
|
||
p50 25.9ms / p99 30.3ms warm reset+import+search on Apple Silicon (per the
|
||
`test/eval-longmemeval.test.ts` perf gate). Per-question cost well under the
|
||
500ms speed gate. 500 questions = ~13s of overhead plus your retrieval and
|
||
LLM latency.
|
||
|
||
## Measuring brain consistency over time (v0.32.6)
|
||
|
||
`gbrain eval suspected-contradictions` is a complementary measurement
|
||
instrument: it samples retrieval results for unmarked semantic
|
||
contradictions (e.g., compiled_truth vs chat content, intra-page chunk
|
||
vs active take). Where LongMemEval measures retrieval correctness on a
|
||
fixed labeled set, the contradiction probe measures how often a real
|
||
brain surfaces conflicting answers.
|
||
|
||
### Recommended nightly cadence
|
||
|
||
```bash
|
||
# Once a day, against your top 50 most-frequent queries:
|
||
gbrain eval suspected-contradictions \
|
||
--queries-file ~/.gbrain/queries.jsonl \
|
||
--top-k 5 \
|
||
--budget-usd 5 \
|
||
--output ~/.gbrain/probe-runs/$(date +%Y-%m-%d).json
|
||
```
|
||
|
||
Persistent cache (`eval_contradictions_cache`) makes re-runs near-zero
|
||
cost until you bump `PROMPT_VERSION`. Trend-track via:
|
||
|
||
```bash
|
||
gbrain eval suspected-contradictions trend --days 30
|
||
```
|
||
|
||
The ASCII bar chart shows total flagged per day. Headline % surfaces in
|
||
`gbrain doctor`'s `contradictions` check with paste-ready resolution
|
||
commands per high-severity finding.
|
||
|
||
### See also
|
||
|
||
- `docs/contradictions.md` — architecture, severity rubric, action criteria.
|
||
- CHANGELOG `## [0.32.6]` — full release notes including the bigger-swing
|
||
decision criteria gated on Wilson CI lower-bound.
|
||
|
||
## v0.40.1.0 Track D — Eval infrastructure
|
||
|
||
Three eval surfaces grew non-trivial capabilities in v0.40.1.0. This section
|
||
covers the dev loop that uses them and the gates they enforce.
|
||
|
||
### `gbrain eval longmemeval --by-type` — per-question-type R@k breakdown
|
||
|
||
LongMemEval has always computed per-question-type recall internally; v0.40.1.0
|
||
surfaces it in machine-readable form. Two additive changes:
|
||
|
||
1. Every per-question JSONL row now includes a `question: string` field so the
|
||
`gbrain eval cross-modal --batch` consumer (below) can read it without
|
||
joining back against the source dataset.
|
||
2. New `--by-type` flag emits a final aggregate line keyed by `question_type`:
|
||
|
||
```json
|
||
{"schema_version": 1, "kind": "by_type_summary",
|
||
"recall_by_type": {"single-session-user": {"hit": 18, "total": 19, "rate": 0.947}},
|
||
"aggregate": {"hit": 110, "total": 120, "rate": 0.917}}
|
||
```
|
||
|
||
**Resume-safe.** When `--resume-from` is the same path as `--output`, the
|
||
summary is rebuilt from the file (each per-row includes `question_type` and
|
||
`recall_hit`) so the final aggregate covers all resumed questions, not just
|
||
this run's slice. The prior summary at the file tail is replaced, not
|
||
appended — a brain that resumes 5 times across a 500-question run ends with
|
||
exactly ONE summary at the tail.
|
||
|
||
**Optional gate.** `--by-type-floor 0.85` exits non-zero when any
|
||
`question_type`'s rate falls below 0.85. Default: informational only.
|
||
|
||
```bash
|
||
# Diagnose per-type ranking quality after a search-touching change.
|
||
gbrain eval longmemeval ~/datasets/longmemeval_s.jsonl \
|
||
--by-type --output /tmp/run.jsonl
|
||
tail -1 /tmp/run.jsonl | jq . # summary line
|
||
|
||
# Strict gate in a CI script.
|
||
gbrain eval longmemeval test/fixtures/longmemeval-mini.jsonl \
|
||
--by-type --by-type-floor 0.80 --output /tmp/run.jsonl
|
||
echo "exit=$?" # 1 if any type fell below 0.80
|
||
```
|
||
|
||
### Hermetic retrieval gate — `test/eval-replay-gate.test.ts`
|
||
|
||
The v0.40.1.0 Track D structural fix for "PRs touching `src/core/search/`
|
||
silently regress retrieval." Replaces the original "replay against captured
|
||
eval_candidates" design (which Codex caught as non-functional in CI — see
|
||
the `v0.41+: contributor-mode CI capture` TODO in `TODOS.md` for the deferred
|
||
real-query version).
|
||
|
||
How it works:
|
||
- Hand-curated qrels fixture at `test/fixtures/eval-baselines/qrels-search.json`
|
||
with PLACEHOLDER names only (no real people / companies per CLAUDE.md privacy
|
||
rule).
|
||
- The test seeds a PGLite engine with synthetic pages whose embeddings are
|
||
basis vectors (the same `basisEmbedding(idx)` pattern as
|
||
`test/e2e/search-quality.test.ts`). No API keys, no DATABASE_URL.
|
||
- For each qrels query, calls `engine.searchVector(basisEmbedding(dim))` and
|
||
computes `top1_match_rate` and `recall@10`. Asserts both meet floors
|
||
(`>= 0.80` and `>= 0.85` by default).
|
||
- Lives in the unit-shard test matrix (`.github/workflows/test.yml`) so it
|
||
runs on every PR via `bun test`, NOT in the E2E fixed-file workflow.
|
||
|
||
#### Refreshing the qrels fixture (the `Why:` discipline, D4)
|
||
|
||
When CI fails because a legitimate ranking change moved expected slugs, the
|
||
fix is to edit `qrels-search.json` directly. **Always include a `Why:` line
|
||
in the commit body** so future maintainers can read the audit trail. Without
|
||
the `Why:`, the gate degrades to a rubber stamp within months. The convention
|
||
is informational (not a commit-hook block), but enforce it in PR review.
|
||
|
||
Example commit body:
|
||
|
||
```
|
||
chore(eval): refresh qrels for new source-boost ordering
|
||
|
||
Why: v0.40.x source-boost now weights originals/ over concepts/, so
|
||
q12 (founder-mode) now correctly surfaces originals/founder-mode-example
|
||
top-1. Manual verification: ran the production query; new ranking is
|
||
clearly better-aligned with the query intent.
|
||
```
|
||
|
||
#### Env-overrides for floors
|
||
|
||
```bash
|
||
GBRAIN_REPLAY_GATE_TOP1_FLOOR=0.85 \
|
||
GBRAIN_REPLAY_GATE_RECALL_FLOOR=0.90 \
|
||
bun test test/eval-replay-gate.test.ts
|
||
```
|
||
|
||
Use to tighten or loosen the gate as the qrels fixture matures.
|
||
|
||
### `gbrain eval cross-modal --batch` — batch quality scoring
|
||
|
||
Single-task cross-modal eval scores one (task, output) pair. Batch mode runs
|
||
the same scoring over an entire LongMemEval JSONL output, with cost guardrails.
|
||
|
||
```bash
|
||
# Step 1: produce LongMemEval hypotheses (real cost: depends on model + N).
|
||
gbrain eval longmemeval ~/datasets/longmemeval_s.jsonl \
|
||
--limit 10 --output /tmp/run.jsonl
|
||
|
||
# Step 2: batch-score those hypotheses (real cost: ~$0.70 for 10 questions,
|
||
# 1 cycle, 3 model slots at default --max-usd 5 budget cap).
|
||
gbrain eval cross-modal --batch /tmp/run.jsonl \
|
||
--limit 10 --cycles 1 --concurrent 3 --max-usd 5 --json
|
||
echo "exit=$?" # 0=all-pass, 1=any-fail, 2=any-error-or-inconclusive
|
||
```
|
||
|
||
**Key behaviors:**
|
||
- Default `--cycles 1` in batch mode (single-task default is 3 in TTY) to bound
|
||
cost. Pass `--cycles 3` to match single-task strictness.
|
||
- `--concurrent 3` runs up to 3 questions in parallel x 3 model slots each =
|
||
9 simultaneous API calls. Below tier-1 rate limits for all three providers.
|
||
- `--max-usd FLOAT` refuses to start if the pre-flight cost estimate exceeds
|
||
the cap, unless `--yes` bypasses (required for non-interactive cron / CI).
|
||
- Filters `kind: "by_type_summary"` rows automatically (the LongMemEval
|
||
`--by-type` summary line is metadata, not a question).
|
||
- `--batch` is mutually exclusive with `--task`; fail-fast usage error if both
|
||
are set.
|
||
- Exit precedence (fail-loud): ERROR > FAIL > INCONCLUSIVE > PASS.
|
||
- Per-question receipts land in a tempdir and are deleted at end of batch; the
|
||
summary inlines per-question verdicts so the audit trail is self-contained.
|
||
|
||
### Nightly cross-modal quality probe (opt-in, autopilot)
|
||
|
||
`src/core/cycle/nightly-quality-probe.ts` ships a phase that runs the longmemeval
|
||
+ cross-modal pipeline once per 24h. **Disabled by default** to avoid surprise
|
||
API spend. Enable per-host:
|
||
|
||
```bash
|
||
gbrain config set autopilot.nightly_quality_probe.enabled true
|
||
gbrain config set autopilot.nightly_quality_probe.max_usd 5.00 # optional override
|
||
```
|
||
|
||
Note: `--phase nightly_quality_probe` wiring into the autopilot scheduler is
|
||
deferred to a v0.41+ follow-up (see TODOS.md). For now the phase is callable
|
||
in isolation; the test harness exercises it via DI stubs.
|
||
|
||
```bash
|
||
# Manual smoke (exercises the path via DI stubs, no real API spend).
|
||
bun test test/nightly-quality-probe.test.ts
|
||
```
|
||
|
||
Observability:
|
||
- `~/.gbrain/audit/quality-probe-YYYY-Www.jsonl` — one event per run with
|
||
outcome (pass / fail / inconclusive / error / budget_exceeded /
|
||
rate_limited / no_embedding_key), pass/fail/inconclusive/error counts,
|
||
est_cost_usd, fixture_sha8. ISO-week rotation (mirrors slug-fallback
|
||
audit).
|
||
- `gbrain doctor` surfaces `nightly_quality_probe_health`:
|
||
- SKIPPED (disabled) — with paste-ready enable command.
|
||
- OK (enabled, no events yet) — autopilot hasn't fired its first run.
|
||
- OK (last 7d all PASS) — with timestamp of latest run.
|
||
- WARN — any FAIL / ERROR / BUDGET_EXCEEDED in the window, with outcome
|
||
counts and the latest run's reason.
|
||
|
||
Real expected cost: ~$0.35 per nightly run (5 questions x 3 slots x 1 cycle
|
||
x ~$0.02/call) ≈ $10.50/month. Worst-case under the default budget cap:
|
||
$150/month. Opt-in default prevents discovering this in your card statement.
|