paper-instruments/feather-train/pdf

PDF documents

A PDF is a fixed-page package rather than a flow document. Existing PDFs generally expose pages, regions, form fields, annotations, overlays, and redactions rather than semantic paragraph reflow. A retained source such as DOCX, PPTX, a build script, or structured data may contain the editable layout model for an authored PDF.

Capabilities

• ReportLab authors PDFs from structured data and explicit page geometry.
• pypdf handles page operations, metadata, forms, and encryption.
• pdf2image uses Poppler to render pages; Pillow handles the resulting images.
• pdfplumber and pypdf extract text and tables.
• Scanned pages without a usable text layer require additional OCR tooling.

A visible rectangle does not remove underlying PDF content. Redaction changes the protected page content; overlays and annotations do not.

References and helpers

• Deterministic PDF quality describes the mechanical

verifier and the limits of extracted-geometry heuristics.

• Content QA describes final-PDF text extraction.

The scripts/ directory contains mechanical checks for openability, page boxes, text layers, extracted content, and selected geometry. The verifier's --strict-layout mode is designed for known generated layouts with predictable geometry, not arbitrary PDFs, scans, assembled pages, forms, redactions, logos, watermarks, or background artwork.