Optionaldpipage-render resolution for engine recognition (default 288, measured;
~400 measured WORSE on ordinary pages — not a "bigger is safer" knob)
Optionalnonever fall back to OCR, even for image-only pages
Optionalocrforce the OCR path even when a text layer exists
Optionalonone event per page as recognition finishes (timing/count only — engine
chatter and the sparse-retry notice arrive via onWarning instead)
Optionalondelivered as each warning happens (also returned, in full, as
warnings on the resolved result)
Optionalonevery recognized word entering the geometry, delivered once after recognition — the severity-1 audit instrument (--debug-words)
Optionalpageemit a ### pN marker before each page's content (output formatting
only — report.pageStats is always populated regardless)
Optionalpagesconvert only this page range (1-based, inclusive); defaults to the whole document
Optionaltuningoverride any subset of the OCR-path geometry defaults — see
DEFAULT_TUNING (src/geometry.ts) for each knob's measured default
Options for pdfToMarkdown.