Oh. I think there’s a misunderstanding here. I don’t ask for a single instance’s opinion on the epistemic correctness of my ideas.

Depending on the problem, I ask for one of the following:

• computational proof if applicable (the skill in question triggers its own audit/code review)
• evidence backed by citations and a rational synthesis session (meta-cognitive skill)
• an adversarial design session (meta-cognitive skill that teaches instances to argue for and against their own beliefs)
• a full triggering of a (possibly distributed) schemata session (another meta-cognitive skill that orchestrates multiple smaller ones, expensive on the token budget and I can’t currently afford it)
• Neckbearding (ditto, but with a Cartesian product matrix of questions vs answers, ultrarationalist style)