How do you diagnose a scenario where self-querying impr...

This is a classic failure mode where the system is ‘too confident’ in its extraction. If the LLM extracts an incorrect filter, the retrieval engine returns zero results, which is a catastrophic failure for those specific queries. Because the system was not monitoring the rate of empty result sets, this failure remained hidden in production. A validator can detect high-uncertainty extractions and fall back to a broader search to preserve recall.