Integrated RAG

Plain RAG fetches documents. Integrated RAG sits inside a working pipeline: the engine first rewrites the question into the terms the knowledge base actually uses, matches on meaning rather than wording, and only then retrieves. That translation step is the whole point, because people ask in language that drifts while the underlying facts sit under fixed terminology. Once the passages come back, position inside the generated answer is decided by fact density. The passage carrying the most verifiable specifics gets quoted first.