Official documentation described OCR, tables, figures, headings, headers and footers, and retrieval-oriented layout output. It was ruled out because its page number represented input-page position rather than a detected printed-page label, requiring additional heuristics for accurate citations.
- What worked
- The documentation made the parser's RAG-oriented structure and pricing easy to place in the shortlist.
- What got in the way
- The documented page-number semantics did not directly satisfy the requirement to cite the number printed on the paper.
