Single committed corpus for regression-testing Codemap — no CODEMAP_ROOT pointing at external apps required for maintainers or CI.
| Piece | Path | Role |
|---|---|---|
| Corpus | minimal/ |
Source tree indexed by goldens, benchmark, agent-eval |
| Golden scenarios | golden/scenarios.json |
Tier A scenario inventory (SQL + recipe ids) |
| Golden snapshots | golden/minimal/ |
Expected query JSON (committed) |
| Capability map | CAPABILITIES.json |
Capability groups → fixture files → goldenScenarios ids |
| Agent eval | scripts/agent-eval/ + probes in scenarios.json |
MCP-on vs traditional arms |
| Benchmark scenarios | benchmark/scenarios.example.json |
Speed comparisons (optional override) |
bun run test:golden # index minimal + compare all scenarios
bun run test:agent-eval # probe/live harness on same corpus
bun run check # unit + golden + agent-eval (CI)
bun scripts/query-golden.ts --update # refresh snapshots after intentional changesbun run test:golden:external remains for consumers validating Codemap against a private checkout (gitignored goldens). It is not part of the Codemap maintainer test bench.
Optional scale (second corpus or rename): docs/roadmap.md · harness map: docs/testing-coverage.md.