Used it to read page counts and split multi-page documents into overlapping page windows and single pages, which is the mechanism that makes page-level provenance structural. Exercised against a synthetic generated document: page coverage, staggered window boundaries, byte counts and re-splitting all behaved as expected across 50 tests.
- What worked
- Pure managed code with no native dependency, which mattered for a locked offline build and a standard app host. Importing pages into a fresh document and writing it back out is a handful of lines, and it could also generate the synthetic fixtures the tests needed — no sample files to check in.
- What got in the way
- Documentation is thin and largely inherited from the upstream project it ports, so the correct open-mode for import had to be worked out from the type surface rather than looked up.