Testing & Mutation Score
Tutela’s correctness is verified two ways: a behavioral test suite, and mutation testing that checks the tests themselves.
Behavioral tests
pnpm test runs 81 assertions against a mock HttpClient layer — no network,
fully deterministic. They cover:
- token fetch, caching, expiry/refresh, and the
Basic/Accept/Content-Typeheaders on the token request; - bearer +
x-api-keyinjection on data requests; - every status → error mapping (400/401/402/404/422/429/5xx, plus the 300 boundary), with normalized field errors;
- transparent 401 refresh-and-retry, and the give-up-after-one path;
- transport errors and malformed-body decode errors;
- curated
effect/Schemadecoding, including a contract-drift case (an out-of-enum value becoming a typedDecodeError).
Mutation testing
pnpm mutation runs StrykerJS over src/core/**.
It makes thousands of tiny edits (“mutants”) to the code and checks that a test
fails for each — a direct measure of whether the suite actually verifies behavior
rather than just executing it.
Score: 93.6%, gated at 90 (CI fails below it). brand, errors, and typed
are at 100%.
The handful of surviving mutants are equivalent — changes with no observable
behavioral difference, which by definition no test can kill (e.g. Context.Tag
identity strings, header literals the mock can’t observe, > → >= boundaries).
They are catalogued rather than hidden; the full discussion lives in
MUTATION.md.
The guiding rule: kill every killable mutant, and document the equivalents. 100% is not a meaningful target — honesty about which mutants are unkillable is.