Extraction was embedded into repository work, not separated from it. Users could review source document, extracted metadata, and structured table data in one surface, turning repository intake into a working system instead of a manual handoff.