NameetP/pdfmux

pdfmux verify — audit ANY extraction engine's output, not just pdfmux's own.

Point it at a source PDF plus an external extraction (Reducto, Mistral OCR, LlamaParse, Docling, an in-house parser) and it re-derives the source with pdfmux's audit pass, then scores every page for coverage, confidence, alignment, hallucination risk, table/heading integrity, and silent drops — the failure where a page has real text but the engine returned nothing while reporting success.

• API: verify_extraction() → per-document manifest; verify_batch() → "M pages silently dropped across N docs"
• CLI: pdfmux verify, with --strict as a CI gate
• MCP: verify_extraction — the 7th tool in pdfmux serve

Also fixed (honesty pass)

• eval/build_fixtures.py shipped a garbage Arabic fixture — PyMuPDF substituted notdef glyphs instead of raising, so the "Arabic" fixture held zero Arabic and the eval reported success anyway. The exact failure pdfmux exists to catch, in our own harness.
• Corrected stale eval numbers published beside that broken harness.
• LICENSING.md claimed the patent-pending method "ships in pdfmux Cloud/Pro" — it shipped nowhere. Corrected to "reserved for".

MIT in full. Full notes: CHANGELOG.md