Retrieval Quality Is a Data Problem

Teams debug retrieval by tuning chunk size for weeks before they check whether the source documents carry the metadata a real query needs.

The metadata you never captured

Document date, author, revision status, and access scope decide more retrieval quality than any chunking strategy. If that metadata was not captured at ingest, no amount of re-chunking recovers it.

Fix the pipeline, not the prompt

We fixed more retrieval-quality complaints by adding a metadata extraction pass at ingest than by changing a single chunking parameter.