Lynx And Hallucination Detection

Operator runs a Lynx-style hallucination / faithfulness check on a RAG answer but passes only the question and the answer, not the retrieved passages the answer was supposed to be grounded in.

Hallucination detection judges whether the answer is supported by the supplied context — it must receive the actual retrieved_context used to generate the answer. Without the real context, a 'faithfulness' verdict is meaningless. Pass the exact passages the generator saw, not a re-retrieval or a su…

A RAG answer faithfully repeats a wrong fact that is present in the retrieved context. The operator treats a 'faithful' (grounded) verdict as 'correct'.

Separate grounding from truth: a faithfulness/hallucination check confirms the answer is supported by the provided context, not that the context itself is correct. Pair faithfulness with a correctness/answer-relevance or reference-based check when ground truth matters, and fix bad source documents …

A long answer mixes supported and unsupported claims. The operator only reads the single top-level pass/fail and cannot tell which sentence hallucinated.

When the detector exposes per-claim or span-level reasoning, surface it so the operator can locate the unsupported span rather than discarding the whole answer. Route the offending span to review; keep supported content. If only a top-level verdict is available, record that limitation. [REQUIRES-VE…