A treatment label that has not earned the right to exclude authority may only rank it.

Treatment labels look like facts. They are measurements, made by editors or by models, and the empirical record shows how contested they are. Reviewing 357 citing relationships that at least one major citator had labeled negative, one study found that all three services agreed negative treatment existed in only 53. Other work found the services' treatment vocabularies applied inconsistently, with categories lacking stable definitions. If the established instruments disagree this often, a label produced by a newer system, human or machine, is not an oracle either.

That has a design consequence, because a label can be used in two ways. It can rank: push an authority down a list, attach a warning, route it to a reader. Or it can exclude: silently remove the authority from what the lawyer sees. Ranking errors are visible and recoverable, because the authority is still there to be read. Exclusion errors are invisible, because nothing in the output reveals what was removed, which makes silent deletion of relevant law among the worst failures in legal research.

The rule follows. A label earns the privilege of exclusion only after a blind audit has measured its error rate on the kind of authority it will exclude, and that rate has been published. Until then it may rank, warn and route, but not delete.

The same principle applies to verifiers that stamp binary verdicts over disputed ground. A verifier that will not publish its measured error rate is asking for the very trust it was built to deny the generator.