# Suite review Mandatory before completing code/test changes, even when everything passes. No retention or deletion quota. ## Review 1. Recover original intent, project contracts, architecture, and test conventions. 2. Inspect new/changed tests + directly overlapping existing tests. Review the affected capability, not the whole repository. 3. Identify adopted techniques; follow the root skill's conditional routes. Combine strategies per code and contract rather than choosing one universal test shape. 4. Ask: - Which intention, plausible bug, or known regression does this protect? - Would it fail for that defect but survive harmless refactoring? - Is its oracle independent? - Is equivalent protection already present, including failure cases? - Which boundaries or cases remain missing? - Is runtime, reliability, diagnosis, and maintenance cost justified? 5. Keep | consolidate | replace | delete | relocate. Remove scaffolding, duplication, tautologies, and obsolete assumptions. Retain useful expensive checks in existing slower tiers. Include the collaboration at risk; a helper-only test cannot prove integration. Keep cheap edge-case matrices where broader tests omit those cases. Do not replace them with one happy-path journey. Integration need not be slow. ## Reject change detectors - Reject assertions pinning incidental wording, private structure, or call choreography. Calling prose a policy, safety instruction, or documentation contract does not make exact phrasing required. Test resource discovery, parsed metadata, links, executable examples, or actual outcomes instead. Regexes, snapshots, and existence checks are not loopholes. - Exact comparisons need intended observable output: protocol values, required literals, or reviewed approved-output baselines. Recorded output alone ≠ correct output. - Keep genuine routing, privacy, authorization, and protocol checks. Judge what an assertion proves, not matcher syntax. ## Replace safely - Diagnose defect | incorrect expectation | authorized behavior change | implementation coupling | flakiness. Difficulty getting green is not a deletion rationale. - Compare scenarios and failure modes, not test counts or touched functions. Where practical, show replacement coverage catches the known defect. Disclose unproven equivalence; no mandatory full mutation campaign. - Deleting most initial tests is fine when protection remains. Unrelated suite weaknesses remain findings, not new scope. ## Finish - Run applicable existing checks and required gates. - Report development evidence, retained protection, actual outcomes, gaps. “All green” describes executed checks, not universal correctness. - Explain material deletion/replacement by protected behavior, not per-test paperwork. - Claim speed improvements only when measured. ## Sources - [Test Desiderata](https://github.com/KentBeck/TestDesiderata) - [Unit Test: boundaries, compile and commit suites](https://martinfowler.com/bliki/UnitTest.html) - [Change-Detector Tests Considered Harmful](https://testing.googleblog.com/2015/01/testing-on-toilet-change-detector-tests.html) - [Fragile Test / Overcoupled Test](http://xunitpatterns.com/Fragile%20Test.html)