What is a hallucination, and how does Centralpoint reduce it?

A hallucination is output that reads as authoritative and has no basis in anything the system retrieved. It is not a bug in the ordinary sense: the model is producing a plausible continuation, which is what it does, and plausibility is indistinguishable from accuracy in the text itself. The defences that work are structural rather than instructional. Centralpoint bounds what the model can draw on by constraining the retrieval surface at index time, so the material available is material the organization approved. Governance-tier skills can require answers to stay within retrieved content and escalate rather than extrapolate where support is thin. And because retrieval draws on governed records with stable identifiers, an answer cites the record it came from and resolves to the version it derived from — so a reader can check in seconds rather than trusting. The AI Interaction Log retains what was retrieved for each execution, which means a disputed claim is verifiable after the fact rather than re-argued.