testing-strategies

Use unit tests for pure functions and isolated business logic, integration tests for API endpoints and database operations, and E2E tests only for critical user journeys like checkout or login. The decision tree maps each feature type to the appropriate test level.

The traditional guideline is 60-70% unit tests, 20-30% integration tests, and 10% E2E tests. Microservices architectures shift toward 60/30/10 with contract tests validating service boundaries.

Use property-based testing with tools like hypothesis, fast-check, or proptest for complex algorithms, parsers, and validators where edge cases are hard to anticipate. Use fixtures for known scenarios and regression tests requiring deterministic data.

No, integration tests should use real ephemeral databases via Docker testcontainers or in-memory SQLite to validate actual queries and transactions. Mock only external APIs using tools like MSW or pytest-httpserver.

Flakiness usually comes from timing issues and unstable selectors. Use Playwright's built-in auto-wait instead of hardcoded sleeps, select elements with data-testid attributes, and add retry logic for network requests.

No, chasing 100% coverage leads to testing trivial code and false confidence. Aim for risk-based targets: 90%+ for critical business logic, 80%+ for API endpoints, and 70-80% overall, optionally validated with mutation testing.