- Tests that reimplement production logic instead of importing and exercising the real code.
- Tests with descriptive names but no meaningful assertions, including `expect(true).toBe(true)`.
- Tests that only assert a mock was called, or assert the exact hardcoded value returned by their own mock.
- Tests that mock internal database, ORM, builder, or service chains so wrong tables, fields, joins, filters, or arguments cannot fail.
- Tests whose fixtures, field names, thresholds, or expected statuses encode behavior different from real producers or the source contract.
- Tests whose names promise inputs or behavior that their setup never exercises.
- Tests that duplicate stronger coverage and add no unique coverage, faster feedback, or better failure diagnosis.
- Tests that copy inventories, manifests, export lists, source imports, or literal strings instead of proving behavior.
- Tests of private predicates, call shapes, or provider-local replays that duplicate shared helper coverage.
- Test-only exports, globals, wrappers, or production paths that exist solely to make tests possible.
- Mocks that implement the behavior being asserted or pretend distinct APIs are equivalent.
- Fixtures that provide receipts, admission, callback ordering, or other outputs the owning production code should generate.
- Persistence tests that assert against a store path the exercised code never writes.
- Tests that assert capability flags instead of actual delivery or acknowledgement.
- Negative controls that pass because an unrelated guard fires or the rejection path is unreachable.
- Tests that sleep or wait on real time, retries, polling, or networks instead of using fake clocks and deterministic boundaries.