Video by NetApp Instaclustr via YouTube

You can catch a hallucination without a human reading every answer. Two cheap, complementary checks did it here: an LLM-as-judge call (does every claim trace back to the context?) and a citation overlap score (do the actual words/numbers in the answer appear in the source text?). Neither check alone is enough — that’s the point of running both.