What the ledger actually finds.

What the audit found

The suite was reported passing for weeks. It was red, at seven failures, and no push had ever been blocked by it.

Why it was invisible

The gate ran through pnpm and Turborepo, which cannot reach a directory with no package.json. The Swift tree had none, so the task silently resolved to nothing and reported success. Three of the seven failures were the repo's own checks for controls that render and do nothing.

What changed

The suite went into the gate, the seven failures were fixed, and an audit of the previous status found the figure that had been reported for weeks had never been printed by any run.

What the audit found

A suite at 526 of 530 passing, with six checks that were structurally incapable of failing on the defect they existed to catch.

Why it was invisible

One read its marker out of a fixture string rather than the system under test. One lost track of which mistake it had found. One guarded an upstream branch that is unreachable. One was blind to any list-shaped field. One compared a basename against an absolute path and had never once executed its body.

What changed

Each was rebuilt to name the mutation it catches. The armed figure and the mutation-naming figure are now published separately, because 526 armed and 309 naming their mutation are two different claims and reporting one number hides that.

What the audit found

Eleven cases claimed a stronger class of evidence than the test underneath them actually earned. The headline coverage figure was therefore too high.

Why it was invisible

A case that asserts an element exists was recorded as proving the behaviour worked. Nobody had lied; the rung a case stands on had never been checked against what the case actually does.

What changed

The ratchet was moved down to the honest number with the arithmetic published beside it, so the figure can only go up from something true. A report that had looked better was replaced with one that was correct.

What the audit found

Sixty-two items measured and broken, two hundred and thirty-one never measured either way, against fifty-seven merged.

Why it was invisible

Work had been merged faster than it had been proved. The board and the repository disagreed, and the board was the optimistic one.

What changed

Nothing yet. This one is included because a page of only clean results is not evidence of a method, it is evidence of selective reporting. The ledger stands at these numbers until the work moves them.