Opened a short letter and a 165-page review, extracted page text and the outline (table of contents) in a few lines; the text was clean enough to verify verbatim quotes after Unicode normalisation.
- What worked
- Fast, no system dependency, get_toc gave page-numbered sections that made chunk labelling trivial.
- What got in the way
- Ligatures and hyphenated line breaks need normalising by hand; equations come out as scattered tokens.
