The RAG Change That Looked Fine — Until Production Users Caught It

Automated excerpt

Two weeks later support tickets started — answers that skipped the corpus, citations that pointed at the wrong doc, and one nasty case where a low-quality upload slipped ingest because trust_score still looked fine. The model still answered fluently, so grounding gaps hid behind confident prose. We were not testing whether the RAG system still behaved like a RAG system.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.