Deterministic Tests

Same code, same result, every run. What this property actually protects is trust in the suite as a whole: one flaky test teaches the team to rerun on red, and from then on the suite’s verdict means nothing. “A test you don’t trust is worse than no test at all” (Martin).

Terminology worth keeping straight: Beck, Google, and GOOS say deterministic (no randomness at all); Meszaros and FIRST say repeatable, the more permissive reading where a seeded RNG, a fixed clock, or a stubbed network all qualify. Repeatable is the bar that matters: no flakes.

Part of Test Desiderata.

Measuring: per-test flakiness rate (Google’s baseline is 0.15%; block merges at 1%), rerun-minutes as a share of CI time, and quarantine debt kept under 2% of the suite.