Merge database integration
Verification Boundary
PostgreSQL, CRM planner and executor are real. Test participant functions deliberately supply blockers or failures; this is not evidence of every external application participant.
What This Suite Proves
Plan decisions, approval material, participant writes, rollback, aliases, durable-job settlement and retained outcomes are asserted through CRM services and PostgreSQL. The scenario bindings identify the assertions inspected in this examination. No control claim is assigned without a check that fails when the control is removed.
What It Does Not Prove
Local database assertions do not establish hosted console behavior, deployed message delivery or the policies of unavailable owner services.
Level and Rationale
A real database is needed to check constraints, row visibility and transaction outcomes. In-memory doubles cannot establish those guarantees.
Real Dependencies and Test Doubles
PostgreSQL, CRM planner and executor are real. Test participant functions deliberately supply blockers or failures; this is not evidence of every external application participant.
Must Not Be Doubled
Keep PostgreSQL constraints and the production persistence or catalog operations real; replacing them would bypass the boundary this selection exists to check.
Fixtures and Isolation
Use the suite fixture workspace and remove its test identities; run database selections sequentially. The test support constructs fixture contexts and test-specific identities. Failure-injection participants belong to the run and must be removed with its fixtures.
Determinism and Retries
The register records no new runtime result. Existing polling and fixture clocks remain part of the authored tests; polling for completion is not a retry of a failed case. Run with at most three workers and diagnose a failure before rerunning it.
How It Runs
The frontmatter records the scoped command and its runner tier. Run the test-database preparation command before invoking this selection.
Coverage Gaps
Mapped scenarios retain their exact execution sources for review; combined multi-source scenarios can still leave an ordered handoff unproved. The capability gap records identify unbound obligations, and the grouped-suite finding records incomplete lifecycle and runtime coverage.