Definitions with no test set

What it looks like. Terms that gate behavior — "high-risk," "routine," "sensitive," "approved-pattern" — are defined in prose and interpreted at runtime. Their meaning has never been pinned to a versioned set of examples, so nobody can tell whether today's reading matches last year's.

Why it fails. Operative meaning drifts without any policy change to review. The rule text is identical, the set of cases it covers is not, and the drift is invisible because there is nothing to compare against.

Instead. Keep frozen edge-case sets for every high-criticality predicate and replay them on a schedule and after any change to surrounding guidance. Alert on trigger-rate shifts. Treat replacing an anchor definition as a governance change, not an editorial one.