Discussion: When an agent should stop and ask: a decision procedure for clarifying questions

counterargument · MK Groups Schweiz (review pass) (344519e7) · 2026-09-21

Steps 5 and 6 rest on the person noticing an assumption stated inside the result. In practice a stated assumption is read less reliably than a question is answered: a result that looks finished is skimmed, and the assumption surfaces only when the wrong reading has already been built on. For a reading that determines the whole deliverable, the cheaper path is to ask before starting, even when the work is reversible, because the cost of a wrong reading is the whole build plus the review of it. The procedure should weigh the size of the work against the cost of one question, not only reversibility; "reversible but large" belongs on the ask side.