do-behavioral-testing

Assert only on user-observable outcomes like rendered text, state transitions, and error messages rather than internal function calls or mock interactions. If a refactor breaks a test while behavior stays the same, the test was coupled to implementation and should be rewritten.

Mock only external boundaries such as network calls, third-party services, and timers that are outside your control. Never mock internal modules or components you own; if mock setup exceeds half the test code, switch to an integration test with real components.

Behavior testing asserts what the user sees, such as "shows error when email is empty", while implementation testing asserts how code works, like which functions were called with what arguments. The former survives refactors; the latter breaks on any internal change.

Use a branch matrix that maps conditions affecting behavior and assign priorities: test all P0 branches independently, test important P1 combinations, and skip logically impossible combos. Four conditions produce 16 combinations but typically only about 8 tests are needed.

Yes, the references include Python equivalents using pytest parametrize for edge case tables, fixtures as test data factories, and respx for mocking HTTP boundaries. The same behavior-first principles apply across JavaScript, TypeScript, and Python.

Tests likely cover only happy paths, which represent roughly 30 percent of code behavior, while error handling and edge cases hold most bugs. Test empty, null, and whitespace inputs separately, cover API failure recovery, and include user chaos scenarios like double-clicks and mid-action refreshes.