The code-health audit found a verification gap: persistence and graph-store tests use fakes and do not execute real PostgreSQL, Neo4j, or Chroma behavior.
Evidence: tests/test_neo4j_store.py:24-82,222-234, tests/test_snapshot_publication.py:41-66,179-197, tests/test_chroma_store.py:10-50, and tests/test_dashboard.py:58-75,872-899. Current tests cannot prove real Cypher acceptance, metadata filters, PostgreSQL serializer/reconnect behavior, or transaction semantics.
Acceptance criteria:
- Default offline tests remain service-free.
- An explicitly marked opt-in suite uses disposable isolated local stores and deterministic fakes for market/model inputs.
- PostgreSQL proves close/recompile/restore and receipt identity/retry behavior.
- Neo4j proves activation failure preserves the prior pointer and immutable conflicts fail closed.
- Chroma proves version/digest filtering and inactive-stage exclusion.
This is a coverage gap, not a confirmed store defect. See the code-health summary and test-coverage-reviewer.md for the scoped prompt.
The code-health audit found a verification gap: persistence and graph-store tests use fakes and do not execute real PostgreSQL, Neo4j, or Chroma behavior.
Evidence:
tests/test_neo4j_store.py:24-82,222-234,tests/test_snapshot_publication.py:41-66,179-197,tests/test_chroma_store.py:10-50, andtests/test_dashboard.py:58-75,872-899. Current tests cannot prove real Cypher acceptance, metadata filters, PostgreSQL serializer/reconnect behavior, or transaction semantics.Acceptance criteria:
This is a coverage gap, not a confirmed store defect. See the code-health summary and
test-coverage-reviewer.mdfor the scoped prompt.