Four new extraction research documents covering permissive error handling with extraction quality signaling (five error classes, circular reference detection, memory limits), PDF/A conformance level guarantees and fast-path optimization (Level A skips OCR and layout heuristics), the complete extraction output schema (span/block/table/NDJSON streaming/ versioning), and per-generator extraction quirks (Word/LibreOffice/ InDesign/LaTeX/Chrome/Ghostscript/scanners). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> |
||
|---|---|---|
| .. | ||
| notes | ||
| plan | ||
| research | ||