Skip to main content

Test pyramid & coverage

Coverage numbers are necessary but not sufficient — they say the tests ran, not that they assert. This standard sets the shape of the suite and makes mutation testing the arbiter of quality.

The pyramid​

Expected ratios, enforced as suite-health checks:

  • Unit — the broad base; fast, isolated, the bulk of assertions.
  • Integration / contract — the middle; real boundaries between components and services.
  • E2E — the thin top; critical user journeys only. Expensive and slower, so kept few and high-value.

A suite that inverts this (heavy E2E, thin unit) is flagged — it's slow, flaky, and hides logic gaps.

Coverage rules​

  • Mandatory diff coverage per PR — changed code must meet the threshold; the merge gate enforces it. Whole-repo coverage is a trend, not a gate.
  • Thresholds are floors, risk-adjusted — higher for security- and money-touching code.

Mutation testing — the real arbiter​

Mutation testing injects deliberate bugs and checks the tests catch them. The mutation score is the measure an agent can't game by writing high-coverage/no-assertion tests. It runs on changed code and gates the merge. This is the single most important defence against AI-generated tests that verify nothing.

Standards referenced: Fowler's test pyramid; Stryker / PIT (mutation testing).