Task
Establish the GATE A foundation assets the roadmap names as preconditions for all breadth work: (1) a versioned golden-corpus of representative repositories with expected ri.v1 fact/evidence contracts, and (2) a capability registry declaring which extractors/constructs are supported, with coverage disclosures.
Rationale
Roadmap §23 W0 (Truth contract) exit criterion: "Every shipped claim maps to tested fact/evidence contracts." §24 GATE A precondition: "Schema conformance and golden-corpus baselines accepted" before any new surfaces or languages are added. Today these are asserted in prose (README "implemented, limited"; REPOSITORY_INTELLIGENCE_V1_RFC.md) but nothing in the repo tracks or enforces them. This gap means the §28 "Trusted output" criterion ("users verify and cite PARTHA evidence") is currently unverifiable at the gate level. Filing this issue makes the gate auditable.
References
- Roadmap §23 W0, §24 GATE A, §28 "Trusted output".
docs/architecture/REPOSITORY_INTELLIGENCE_V1_RFC.md (ri.v1 contract).docs/architecture/REPOSITORY_INTELLIGENCE.md.- Existing backend test suite (
apps/backend/tests) — extend rather than duplicate.
Implementation Notes
- Golden corpus: a small, licensed set of fixture repos under
tests/golden/ with committed expected ri.v1 snapshots or contract assertions; a CI job that diffs extractor output against them and fails on silent drift.
- Capability registry: a single source-of-truth manifest (e.g.
app/intelligence/registry.yaml or a module) listing supported languages/constructs + support level, surfaced by the docs/README "implemented, limited" claims so they cannot drift from code.
- Non-goals: no new product surface, no new language support — this is foundation only.
- Must respect AGENTS.md: no second parser; reuse
app/intelligence/ extraction.
Acceptance Criteria
Task
Establish the GATE A foundation assets the roadmap names as preconditions for all breadth work: (1) a versioned golden-corpus of representative repositories with expected ri.v1 fact/evidence contracts, and (2) a capability registry declaring which extractors/constructs are supported, with coverage disclosures.
Rationale
Roadmap §23 W0 (Truth contract) exit criterion: "Every shipped claim maps to tested fact/evidence contracts." §24 GATE A precondition: "Schema conformance and golden-corpus baselines accepted" before any new surfaces or languages are added. Today these are asserted in prose (README "implemented, limited"; REPOSITORY_INTELLIGENCE_V1_RFC.md) but nothing in the repo tracks or enforces them. This gap means the §28 "Trusted output" criterion ("users verify and cite PARTHA evidence") is currently unverifiable at the gate level. Filing this issue makes the gate auditable.
References
docs/architecture/REPOSITORY_INTELLIGENCE_V1_RFC.md(ri.v1 contract).docs/architecture/REPOSITORY_INTELLIGENCE.md.apps/backend/tests) — extend rather than duplicate.Implementation Notes
tests/golden/with committed expected ri.v1 snapshots or contract assertions; a CI job that diffs extractor output against them and fails on silent drift.app/intelligence/registry.yamlor a module) listing supported languages/constructs + support level, surfaced by the docs/README "implemented, limited" claims so they cannot drift from code.app/intelligence/extraction.Acceptance Criteria