testing-strategies

Follow the testing pyramid: roughly 70% fast unit tests for logic and edge cases, 20% integration tests for API contracts and data flow, and 10% E2E tests for critical user journeys. Avoid the inverted pyramid, which causes slow and flaky CI.

Mock external HTTP APIs, databases in unit tests, time, file system, and random values because they are slow or non-deterministic. Never mock your own business logic or pure functions, since that tests the mocks instead of the code.

Coverage between 50-70% is reasonable for most projects, while 70-85% shows diminishing returns. The more meaningful metric is coverage of changed code in each pull request, since 100% coverage without real assertions is useless.

Flaky tests usually stem from timing races, environment differences between CI and local machines, test pollution from shared state, or hardcoded timeouts. Fix them with proper waits, pinned versions, isolated setup/teardown, and event-based synchronization.

TDD works best for well-understood requirements where the red-green-refactor cycle guides incremental design. For exploratory code where the design is still unstable, write tests after the implementation settles to avoid constant test rewrites.