[felipesauer]

The first outside contribution went red on an internal count a guard keeps by hand, which its author could neither see nor run.

Four guards keep a count per file (the import clauses per file, the commands each page hands over in two guards, the passages of the overwrite recipe per page). Now:

• a count that moved fails with the file and line of the ledger and the number to write, plus the command that writes it;
• pnpm vitest run <the guard's file> -u rewrites exactly those numbers in place, the way -u rewrites a golden;
• a file with no line, a line with no file and a row that gains or loses a kind of count still fail: someone lists those on purpose, and -u never lets them in.

Kept by hand on purpose: the ledgers of reasons (the unwitnessed files, the allowed escapes, the instrument list of the rulers guard), whose point is the sentence a person writes.

CONTRIBUTING.md says a red count is fixed by the command it names and that a maintainer does it when the contributor cannot.

[felipesauer]

Review verdict: approve (c5fa848). Each counting guard now fails with the file, the line and the number to write, and pnpm vitest run <file> -u rewrites only that number. The update mode is inert in CI, and a new file still has to be listed by a person. A test holds the detection path, so a plain run never rewrites a file. CI is green on main merged in.