If there were specific criteria that I could share in an HN comment, they would be used for hiring and immediately collapse under Goodhart’s law. Your question is almost tautologically bad - If I were able to answer it in a brief HN post, my answer would be bullshit.

You can accurately probe competence by asking people questions about their thought process, their current hypotheses, etc. If there are $BIGNUMBER tests for competence, you only need to ask $SMALLNUMBER of them to make an accurate determination as long as the person being asked doesn’t know them ahead of time! This is the same trick many zero-knowledge proof algorithms use.