The keeper metaphor came from Claude and I loved it too. Left it verbatim. Good point on flagging specifics for fact checking. I've run into many hallucinated 'facts' in my work with AI. I prompt to 'red team' the ideas we work on frequently and have standing instructions that attempt to counterbalance sycophancy and failure by coherence. I like to feed whole chats to other LLMs to see what bones they choose to pick.