Firecrawl pdf-inspector: Quality-Aware PDF Ingestion

Overview

Firecrawl's pdf-inspector treats PDF ingestion as a page-level evidence problem: native parsing preserves font, geometry, and layout signals; quality checks distinguish trustworthy text from garbled or incomplete coverage; selective OCR adds visual evidence only where needed; fusion and provenance determine what can be published or escalated. The series moves from PDF internals through decoding, layout, routing, and deployment to a conditional policy. It supports defensible local Markdown, not universal completeness.

Learning outcomes

Native parsing preserves positioned evidence, but readable output does not prove complete page coverage.
Font decoding, geometry, and coverage are separate checks; strength on one cannot erase failure on another.
Selective OCR should target affected pages or regions, then compare candidates instead of automatically replacing native text.
Provenance and hosted escalation keep weak, incomplete, or unavailable local results from looking complete.

Keywords

pdf-inspectorPDFExtractionOCRLayoutGeometryFontsProvenanceMarkdownFusion

Chapters

pdf-inspector: When PDF Extraction Fails

Classify pages, preserve uncertainty, and escalate when native evidence runs out

14 min · Classify pages, preserve uncertainty, and escalate when native evidence runs out

14 min

pdf-inspector: PDF Content Streams

How parser state turns painted operations into positioned evidence

13 min · How parser state turns painted operations into positioned evidence

13 min

pdf-inspector: Font Mapping and Text Quality

How Unicode recovery exposes garbling before OCR routing

13 min · How Unicode recovery exposes garbling before OCR routing

13 min

pdf-inspector: Native Evidence Checks

Test presence, decoding, placement, and coverage before escalation

15 min · Test presence, decoding, placement, and coverage before escalation

15 min

pdf-inspector: Layout-Aware Markdown

Recover reading order and tables without overtrusting geometry

18 min · Recover reading order and tables without overtrusting geometry

18 min

pdf-inspector: Selective OCR Routing

Choose pages, preserve transforms, and load vision only when needed

15 min · Choose pages, preserve transforms, and load vision only when needed

15 min

pdf-inspector: OCR Fusion and Provenance

Choose native, OCR, fused, or hosted output by evidence

17 min · Choose native, OCR, fused, or hosted output by evidence

17 min

pdf-inspector: Production Ingestion Policy

Choose native, OCR, fusion, or escalation by page evidence

14 min · Choose native, OCR, fusion, or escalation by page evidence

14 min

Preview every chapter · Unlock full text, audio, and video from chapter 4

Plans

Ch.1-3 FreeCh.4-8 Premium

FAQ

What does pdf-inspector actually do?

pdf-inspector classifies PDF pages, extracts native structure, reconstructs layout, and routes weak pages to selective OCR.

When should a PDF page use OCR?

A page should use OCR when native evidence is missing, garbled, vectorized, or visually incomplete; clean pages can stay native.

Can native PDF extraction prove that a page is complete?

No. Native extraction can report recovered structure and quality signals, but an unflagged result does not prove that every visible word was represented or correctly interpreted.

How should native text and OCR be combined?

Compare quality, overlap, confidence, and new coverage: keep stronger native text, fuse material additions, use OCR when native is rejected, and recommend hosted processing when both remain weak.