Expand description
Stage 2 — extract: turn the verbatim source into a canonical structured representation: plain text (the cheap full-text surface) and canonical HTML (the structure-bearing surface that the CML layer annotates).
The default extractor is pure-Rust pdf-extract so the tool always builds
and runs with no system libraries. It produces faithful text but no
per-page markers or MathML; those are the quality knob delivered by heavier
extractors (PyMuPDF / nougat) wired in behind the extractor field later.
HTML here is built deterministically from the text by paragraph/heading
heuristics — honest structure, not invented content.
Structs§
- Extracted
- Result of extraction — what becomes the
derived/assets.
Functions§
- extract
- Extracts.