The engine matches specific grounded response segments to the retrieved source page content using an LLM evaluator to detect survived facts, omissions, and hallucinations, yielding a per-URL fidelity score.
Referenced by
