A three-category test census across seven repositories

What was wrong before this entry: the homepage carried a single automated-tests figure with no category, and each system's own manual witnessed a runner count that never reached the site. The three different things a serious evaluator asks about — how many tests, how many red-team attacks, how many audits — were flattened into one number.

The census now counts three categories per repository and publishes them where they belong: the homepage carries the aggregate, and each brochure carries its own system's three numbers. The methodology is recursive — no directory-skipping search — and the test category carries the runner count each system's manual states at the bytes, not a declaration count guessed from file names.

The rule that survives this entry is the one already in the record: recursive census or no number. A count that stops one directory short has already produced one wrong repository count in this project, and it will again if it is allowed to.