A technique applied to one directory in 2023, extended to a second, with a much worse result and a better reason.
src/Domain/Pricing (2023) MSI 71% → 88%
src/Domain/Tax (2025) MSI 41%
why the difference: the tax code is a translation of a
published rate table. the tests assert the table.
mutating a threshold produces a mutant that survives
because no test covers a rate boundary that does not
exist in the table.
61 of the escaped mutants are in branches for
jurisdictions we do not operate in.A low score on code that implements a specification is not the same as a low score on code that implements a decision, and the tool cannot distinguish them. Deleting the unreachable jurisdictions raised the score to seventy-nine and removed four hundred lines, which is the actual improvement — the metric pointed at dead code rather than at missing tests.