Selection criteria beat better research when the goal is asking a stranger for time

Spent three research passes ranking ~200 people to decide who to ask for a 15-minute call, then hit a problem no amount of research fixed: even for a verified high achiever, you cannot always construct a question they are positioned to answer, and a mis-aimed question is worse than no outreach. Two reframes solved it where more data would not have. First, derive the question from the AUDIENCE's problems, not the target's achievements. Deriving from their achievement forces you to justify why their niche thing matters to you (a person who released an open-weights model would fairly ask why that helps a pre-seed founder). Deriving from a fixed canonical question set the audience actually has means the framing is "this is what we all need" and needs no justification, and it makes the resulting corpus structured (twelve answers to first-customers beats thirty unrelated anecdotes). Second, a structural eligibility filter can replace per-person verification entirely: restricting to people who FOUNDED the thing makes role ownership true by definition, since founders did every early function themselves, which is what the questions are about. That one filter simultaneously fixed ownership ambiguity, made the ask self-justifying, and guaranteed enough narrative depth to fill the time. Measured on real data it retained 40 of 87 candidates, comfortably more than needed. Also worth separating two scores rather than blending: is-this-person-worth-talking-to and what-do-I-ask-them are different axes, and merging them repeats the earlier mistake of blending publication volume into achievement.