mirror of
https://github.com/garrytan/gbrain.git
synced 2026-07-31 04:07:52 +00:00
Garry Tan
18baca6629
Adds the v1→v2 contract boundary for BrainBench. 6 JSON schemas at
eval/schemas/ pin the shape of every artifact a stack must emit to be
scorable: corpus-manifest, public-probe (PublicQuery with gold stripped),
tool-schema (12 read + 3 dry_run tools, 32K tool-output cap), transcript,
scorecard (N ∈ {1, 5, 10}), evidence-contract (structured judge input).
8 gold file templates at eval/data/gold/ scaffold the sealed qrels,
contradictions, poison items, and citation labels. Empty-but-valid
skeletons; Day 3b fills them with real content once the amara-life-v1
corpus generates.
48 tests validate schema syntax, $schema/$id/title/type headers,
round-trip stability, and cross-schema coherence (new Page types in
manifest enum, tool counts, token cap, N enum).
When v2 ports to Python + Inspect AI + Docker, these schemas are the
boundary. Same fixtures, same tool contracts, zero rework.
16 lines
641 B
JSON
16 lines
641 B
JSON
{
|
|
"version": 1,
|
|
"_comment": "Cat 5 gold. 100 claims sampled from a seeded gbrain-adapter brain-export, each labeled `supported | unsupported | over-generalized` against source material. Threshold >0.90 citation_accuracy (per design-doc METRICS.md).",
|
|
"claims": [
|
|
{
|
|
"_example": "true",
|
|
"id": "claim-001",
|
|
"source_page": "people/jordan-park",
|
|
"claim_text": "Jordan Park co-founded NovaMind in 2024.",
|
|
"expected_label": "supported",
|
|
"expected_evidence": ["doc/novamind-investor-update", "meeting/mtg-0003"],
|
|
"reason": "Both sources explicitly name Jordan as co-founder with date"
|
|
}
|
|
]
|
|
}
|