Research Rigor: a three-phase protocol against hallucination

The objective is simple: don’t let an AI sound confident about something it hasn’t actually checked. I load it explicitly whenever a task needs to be right, not just plausible: market sizing, competitive claims, anything going into a document that will be scrutinized. It runs three phases every time: state every assumption and ask clarifying questions before answering, research and cite sources while answering, then run a skeptical self-critique pass before delivering.