[emmanuel-a-otchere]

Summary

• Implement src/contracts/semantic-integrity.ts:
  • TC-SEM-01 URN grammar (urn:source-engineering:<type>:<id> on every cots/source/decision/evidence/claim/constraint identifier)
  • TC-SEM-02 evidence-chain resolution: every source_id resolves to a source artifact, every claim_id to a claim, every claim evidence reference to an evidence item, every decision_id/decision_record_id to a DecisionRecord. No dangling references
  • TC-SEM-03 temporal ordering: no evidence issued after produced_at, no validation before production. Source expiry remains TC-CI-03's warning; TC-SEM-03 only fails impossible orderings
  • TC-RC-01..05 (RFC-0002): model-generated evidence items warned under admitted policies, bounded recursion requires a max_levels bound (bound > 1 warns), unbounded recursion warns, terminal policy violated by a model-generated item fails, legacy pre-RFC-0002 COTSs warn
• Add the RFC-0002 optional fields to src/types.ts (source_type, chain_termination)
• Wire the four validators into the runner
• Add 9 fixtures and tests/semantic.test.ts (14 tests incl. two end-to-end runner tests)

The suite now proves the COTS is trustworthy, not just present: this operationalizes the first case study's Q3.2 finding (a COTS citing itself).

Verification

• npx vitest run: 37/37 tests pass
• npx tsc --noEmit: clean