64 / 67 · 06 Observability-Driven Testing · Correlating Test Results with Production Metrics← prev⊞ allnext →☰ Read as one page
10.4Three Categories of Tests
Load-Bearing Tests
Tests whose failure predicts production incidents. These are your most valuable tests.
How to identify them:
- They failed before a related production incident
- They cover critical business logic or integration points
- Disabling them would increase incident risk
What to do: Protect these tests. Never skip them for speed. Prioritize fixing them when they flake.
Coverage Gap Tests (Missing Tests)
Tests that should exist but do not. Identified when a production incident occurs in a component with no test coverage for that failure mode.
How to identify them:
- Run the correlation analysis after every incident
- Map incidents to components and check test coverage
What to do: After every postmortem, add at least one test that would have caught the incident.
Dead Weight Tests
Tests that always pass, never catch bugs, and provide no meaningful signal.
How to identify them:
- Last failure was more than 6 months ago
- They test trivial logic (obvious constants, simple getters)
- They duplicate coverage with other tests
- No production incident has ever related to their component
What to do: Consider retiring or consolidating them. A smaller, focused test suite is more valuable than a large, noisy one.