For tests, prefer deterministic, intention-revealing assertions and keep runtime tests and type-only tests strictly separated. Apply: 1) Use the right matcher for the claim - Identity: use `toBe` instead of deep equality when referential identity matters. - Async rejections: prefer matcher-based assertions over try/catch. 2) Make timer-based tests fast and stable - If you must use timers, prefer fake timers and advance/run them deterministically. - Keep real timer durations extremely short. 3) Don’t mix runtime and type-only tests - Put compile-time checks into `*.test-d.tsx` and use `expectTypeOf(...).toEqualTypeOf(...)`. - In `*.test-d.tsx`, don’t rely on runtime behavior inside `it` blocks. 4) Ensure the test can prove which code path executed - When a hook can call different `queryFn`s (e.g., prefetch vs render), use distinct spies and assert call counts. Clear mocks after setup calls. Example (async rejection + timers + type-only): ```ts // runtime test await expect(promise).rejects.toBe(testError) // timer test (prefer fake timers) vi.useFakeTimers() vi.runAllTimersAsync() expect(spy).toHaveBeenCalledTimes(0) // type-only test in *.test-d.tsx expectTypeOf(result4[0]).toEqualTypeOf>() ``` This reduces brittle/flaky tests, improves signal-to-noise, and guarantees type coverage lives where it’s actually executed by the tooling.