Non-trivial changes SHOULD begin with a test.
A test SHOULD fail before implementation begins whenever practical (red → green → refactor).
Implementation SHOULD follow the red → green → refactor cycle.
Implement the smallest change required to make the test pass.
Refactoring MUST happen only after tests pass and MUST NOT change behavior.
Tests MUST act as executable specifications of behavior, not merely coverage.
Tests MUST assert observable behavior, not implementation details.
Bug fixes SHOULD include a regression test when feasible.
Code changes SHOULD include corresponding test updates when behavior changes.
Tests MUST be deterministic and isolated from uncontrolled inputs (for example network, time, or randomness).
Tests SHOULD avoid reliance on external systems unless explicitly intended.
External dependencies SHOULD be controlled appropriately.
Tests SHOULD target the lowest meaningful level.
Integration and end-to-end tests SHOULD complement lower-level tests where needed.
Tests MUST NOT be weakened or removed to satisfy incorrect implementations.
Tests SHOULD fail for a single clear reason.
When tests cannot be added, this MUST be rare and explicitly justified, including:
- why testing is not feasible
- what risk remains
- how the change was validated
Skipping TDD for non-trivial changes REQUIRES explicit justification.

Scope

Non-trivial changes include:
- Bug fixes
- New features
- Behavior changes
- Data transformations
- Business logic or query changes
Excluded:
- Pure formatting changes
- Renames without behavior change
- Comment or documentation updates

TDD MAY be relaxed for:
- Rapid prototyping or spike solutions
- Exploratory coding where requirements are unclear
- Pure UI layout or styling changes

writing-tests-first