blog: the auditor made the mistake it audits - #872

Eleventh post in the CI-gates series, and the first about the tool rather than the gates.

A mutation tool took the gate suite from four gates with no negative control to none, and from twenty surviving mutants to zero. In the same week it made, three separate times, the exact defect it exists to find — a checking function covered by tests, and the wiring from it to the answer not.

• --invert shipped, merged, published a number, and never ran. The enum variant was declared and documented, invert_sites() was written and unit-tested, and nothing joined them. Ten unit tests passed with the bug present; each exercised one function or the other, none crossed between them. What made it stand was that the answer was plausible.
• The boundary operator's first run counted prose. Quote state was tracked per line, so every > inside a multi-line docstring became a site — 8 of 13 gates, corrected to 5 once triple quotes carried across lines.
• The equivalence marker took marker + 2 lines against a fifteen-line proof, and named a line in the middle of its own explanation.

And the counter-case, which is the useful one: a limitation whose every clause was true and whose conclusion did not follow, because it reasoned entirely inside the frame of the helper that file happened to be built around.

Both locales. Four receipts. Four open questions, including that "no mutant survives" is not "the gates are correct" — the operators are narrow by construction, and whether the gates check the right properties is a question none of them asks.

Verified in the built bundle before opening: slug present in Blog-*.js, both titles present, three load-bearing numbers present.