Groundedness

A grounded statement can be traced to a passage the system retrieved; an ungrounded one came from the model's training and may be true, outdated or invented. The distinction is invisible in the output, which is why fluent answers are trusted more than they should be. Measuring groundedness means checking generated claims against the retrieved context and flagging those with no support. In regulated settings the requirement often goes further: an answer that cannot be grounded should decline rather than proceed, because a plausible unsupported answer is worse than an admission that the corpus does not cover the question.

Because Centralpoint retrieves from governed records with stable identifiers and retains what was retrieved per execution, the supporting passages for an answer are available for comparison rather than reconstructed afterwards. Governance-tier skills can require that answers stay within retrieved material and escalate rather than extrapolate.