Three tests you run on any question before you ask it — whether the source is a witness, an interviewee, or an AI. Sharp = open AND specific AND clean.