Establish a passing relevant baseline and a protection question.
Scope production files/functions/lines/mutants using supported selectors.
Select relevant tests through existing per-test coverage mapping or small test groups.
Mutant scope ≠ test selection; omitting a legitimate detecting test can mislead.
Reuse valid caches/history.
Inspect useful survivors, improve behavioral protection, rerun affected scope.
Improve high-level tests when appropriate, not one test per mutant.
Survivor ≠ bug.
Distinguish surviving, uncovered, equivalent, invalid, timed-out, and infrastructure outcomes using runner definitions.
Do not encode mutation operators as requirements or chase a perfect score.
Cache discipline
Check tracked inputs: goldens/snapshots, fixtures, dependencies, environment, config, helpers, and test deletions may invalidate history.
Relevant untracked input changed → force fresh affected-scope analysis through supported automation.
Unchanged production source alone does not justify reusing a kill.
Report fresh | reused | partial | stale evidence and selected scope.
Combined reports may include historical results outside the new run.
No supported optimization → bounded run or defer with cost/limits disclosed.
No automatic infrastructure repair, new gates, broad campaign, erased survivors, or weakened thresholds/operators.
# Mutation testing
Also: mutation analysis.
Mutants are changed program variants; mutation fuzzing is different.
Coverage asks “did it run?”
Mutation asks “did selected tests detect selected changes?”
Neither establishes overall correctness.
## Workflow
1. Establish a passing relevant baseline and a protection question.
2. Scope production files/functions/lines/mutants using supported selectors.
3. Select relevant tests through existing per-test coverage mapping or small test groups.
Mutant scope ≠ test selection; omitting a legitimate detecting test can mislead.
4. Reuse valid caches/history.
5. Inspect useful survivors, improve behavioral protection, rerun affected scope.
Improve high-level tests when appropriate, not one test per mutant.
Survivor ≠ bug.
Distinguish surviving, uncovered, equivalent, invalid, timed-out, and infrastructure outcomes using runner definitions.
Do not encode mutation operators as requirements or chase a perfect score.
## Cache discipline
- Check tracked inputs: goldens/snapshots, fixtures, dependencies, environment, config, helpers, and test deletions may invalidate history.
- Relevant untracked input changed → force fresh affected-scope analysis through supported automation.
Unchanged production source alone does not justify reusing a kill.
- Report fresh | reused | partial | stale evidence and selected scope.
Combined reports may include historical results outside the new run.
- No supported optimization → bounded run or defer with cost/limits disclosed.
No automatic infrastructure repair, new gates, broad campaign, erased survivors, or weakened thresholds/operators.
## Sources
- [Stryker: incremental runs and invalidation limits](https://stryker-mutator.io/docs/stryker-js/incremental/)
- [Stryker: coverage-based test selection](https://stryker-mutator.io/docs/stryker-js/configuration/#coverageanalysis-string)
- [PIT: history and dependency-tracking limits](https://pitest.org/quickstart/incremental_analysis/)