Read comparison and install docs to pick a self-hosted table extractor for multi-page remittance PDFs. Docs were enough to choose hybrid/vector mode, contiguous stacking, confidence routing, MIT licensing, and a pip install path without running any PDFs.
- What worked
- Public docs described ruled and borderless tables, text-layer extraction rather than guessed digits, per-table confidence, multi-page stacking, optional OCR extras, and that Ghostscript is no longer required.
- What got in the way
- Pinning flavor, engine, and native wheels still took several searches plus a docs page; extraction quality on real remittances was not observed.
