Traditional RAG systems treat documents as opaque text blobs — they chunk blindly, embed vaguely, and retrieve by fuzzy similarity. You get an answer but can'"'"'t trace it back to the source.