platform/ci/tax_year_staleness.py lists calculators that cite a tax authority but show no figure for the current year. A red-team run seeded ten small faults into it (changing >= to >, and to or, a limit of 40 to 41). The existing tests caught two. Eight escaped.
The gap was plain: the tests only exercised scan(), never main(). Nothing checked the "at least three authority mentions" threshold, that current-year products are left out, the default --limit of 40, the fixture directory exclusion, or that a year two ahead is ignored while next year counts.
Five new tests cover those, and a re-run of the same ten mutants now catches all of them. Across the run's three targets (22 mutants), the catch rate went from 64% to 100%. The other two files, check_social_proof.py and figureless_projects.py, already caught every mutant.
This is a report, not a gate, so a silent regression would not have broken CI. It would just have quietly stopped telling us which calculators carry last year's brackets.