Phase 5C — KB Unit Test Coverage Expansion (close 32-vs-12 parity gap with MC)

Context

Sub of Phase 5 Epic #11643 (meta-Epic #11624).

Cross-cutting; can start any time. Audits + backfills KB unit test coverage gaps to approach MC parity.

The Problem

Empirical parity gap (V-B-A grounded; calibrated 2026-05-19T13:42Z per GPT epic-review on #11643 — initial path-glob was too broad):

Path-scoped counts (canonical test/playwright/unit/ai/services/):

• memory-core/*.spec.mjs (incl. subtree): 23 specs
• knowledge-base/*.spec.mjs (incl. subtree): 17 specs
• KB has ~74% of MC's unit test density

Initial framing (find -path "*memory-core*" → 32 vs *knowledge-base* → 12) swept broader subtrees including ai/daemons/services/ and incorrect path roots. The path-scoped count is the canonical metric.

While the 6-spec delta is smaller than initially framed, substantial coverage gaps likely still exist in:

• DatabaseService (570 lines — orchestrator; backup/restore methods; lifecycle)
• VectorService (308 lines — embedding pipeline; content-hash delta; #10572 volume gate behavior)
• QueryService (222 lines — Chroma query construction)
• SearchService (170 lines — file hydration path)
• HealthService (374 lines — health surface)
• KBRecorderService (450 lines — telemetry pipeline; Memory Core SQLite integration)
• ChromaManager (142 lines — Chroma client lifecycle)

Operator framing: "and /unit-tests too" — explicit directive to close the gap.

The Fix

1. Audit current KB unit coverage — for each KB service, enumerate public methods + identify which have unit-test coverage vs which don't. Produce gap report.
2. Prioritize backfill targets — focus on:
   • Methods that will be MODIFIED in Phase 0/1, 2, 4 (regression-protection priority)
   • Methods with non-trivial logic (error paths, edge cases, state transitions)
   • Methods that interact with shared substrate (Chroma, Memory Core SQLite, AgentIdentity)
3. Backfill targeted unit tests — author new .spec.mjs files OR extend existing specs
4. Track parity progress — measure KB spec count after backfill via path-scoped count of test/playwright/unit/ai/services/knowledge-base/**/*.spec.mjs; target ≥20 specs (~87% of MC's 23 path-scoped count; pragmatic V1 target)

Acceptance Criteria

• KB unit-coverage audit report committed to test/playwright/unit/ai/services/knowledge-base/COVERAGE_AUDIT.md (or similar; format TBD by implementer)
• Backfill targets prioritized + documented; rationale per service
• At least 3 NEW unit spec files added (path-scoped 17 → ≥20)
• Coverage emphasizes Phase 0/1, 2, 4 modification surfaces (regression-protection priority)
• All new specs pass via npm run test-unit (NOT npx playwright)
• No flakiness introduced (existing test stability preserved)
• Optional: coverage tool report (istanbul / c8) showing pre/post backfill delta

Out of Scope

• Integration tests (Phase 5A/B own that)
• E2E tests (Phase 2F owns that)
• Closing the FULL 32-spec gap if pragmatic 25+ is achievable in single sub (further backfill can spawn follow-up ticket if needed)
• Refactoring tested code to be more-testable (testing AS-IS; refactoring is separate concern)
• Performance regression testing

Avoided Traps

Related

• Parent: #11643
• Empirical anchor: find test/playwright/unit/ai -path "*memory-core*" -name "*.spec.mjs" | wc -l = 32; KB equivalent = 12 (verified 2026-05-19)
• Reference patterns: test/playwright/unit/ai/memory-core/*.spec.mjs (MC unit test conventions)

Origin Session ID

7360e917-1733-4cdd-a6f3-5ac51c34b838

Handoff Retrieval Hints

• Start with the audit phase — gap report INFORMS prioritization; don't backfill blindly
• Memory anchor: feedback_npx_bypass_test_isolation.md — npm run test-unit not npx playwright
• Memory anchor: feedback_mcp_test_location.md — KB unit tests under test/playwright/unit/ai/knowledge-base/
• KBRecorderService.mjs is high-priority backfill target — 450 lines + extends into Phase 4A
• VectorService.mjs:188-274 is high-priority (Phase 0/1C modification surface)
• QueryService.mjs:116-128 is high-priority (Phase 0/1D modification surface)