I've been working on a problem: AI agents confidently claim to understand things they don't, make the same mistakes across sessions, and have no awareness of their own knowledge gaps. Empirica is my attempt at a solution – a "cogniti…