hardening-tests

Hardening tests (SDET)

A passing suite isn't a good suite. Tests can pass while asserting nothing, miss whole branches, or flake under load. This skill attacks the suite the way reality will, and strengthens whatever gives. It never adds product code — its only output is stronger tests.

What to check (in rough priority order)

1. Coverage gaps — which branches/packages are untested. Use the project's coverage tooling; treat the floor as a floor, not a target. Find the untested edge, not the %.
2. Assertion strength — would the test fail if the behavior were wrong? The sharpest tool here is mutation testing: deliberately break the code and see if a test catches it. Tests that pass against broken code are guarding nothing. See references/mutation-testing.md.
3. Flakes & non-determinism — run hot, under -race, many times. A test that fails 1 in 20 will fail in CI at the worst moment. See references/flake-hunting.md.
4. Missing edge / property / fuzz — invariants deserve property/fuzz tests, not just examples (pricing totality is already fuzzed here; extend the pattern). See references/assertion-strength.md.
5. Unguarded fixed bugs — cross-check docs/REGRESSIONS.md: every fixed bug should name a guard. Missing ones are the highest-value tests to add.

Scripts (deterministic, repetitive — run them)

scripts/flake-hunt.sh ./internal/web 20   # run a package N times under -race; report any failure
scripts/mutation-run.sh ./internal/telemetry  # run the mutation tester if installed; else explain setup

Boundaries

practicing-tdd and authoring-tests create tests; this skill attacks and strengthens them. If hardening reveals a real product bug (not just a weak test), file it via tracking-issues and let TDD fix it — don't fix product code from here.

Why mutation testing earns its keep

Coverage says a line ran; mutation testing says a line is checked. A suite can have 90% coverage and still let you delete a ! without a single failure. Mutation testing is how you find those silent passes — it's the closest thing to an objective test-quality score.