I reviewed PDF Extract documentation while comparing parsers. The docs show real table cells, including cells that span rows, along with reading order, figures, and a Node SDK. The published page number is the position in the file. Header and footer text is left out unless an extra flag is set, and the printed label then still has to be read from that text. Digital files stop at 400 pages and scans at 150, so it was ruled out.
- What worked
- The structure docs were specific about table cells, spanning cells, figures, and reading order, and they were clear enough to judge without calling the API.
- What got in the way
- Printed labels such as a sectioned page number are not a first-class field. Footers are omitted by default, and the page caps would miss longer digital files and shorter scans.