Dataset · 22 workflows
Browser PDF tools accuracy benchmark
See the stress case, independent validation method, exact result, and output checksum for merge, split, compression, security, OCR, conversion, and image workflows.
Open PDF research
PDF benchmarks and reference data you can verify
Most converter claims stop at “fast” or “high quality.” This library publishes exact inputs, measured outputs, formulas, limitations, and downloadable data so the claims can be checked independently.
Use the research to choose settings or check a result
Each resource pairs original measurements with downloadable inputs, outputs, formulas, or validation steps so you can inspect the evidence yourself.
Calculator · 64-row open dataset
PDF to PNG DPI and pixel-size chart
Calculate export dimensions, megapixels, and uncompressed memory for Letter, A4, Legal, or Tabloid pages from 36 to 1,200 DPI.
Generator · safe test fixture
Sample PDF generator for testing tools
Create a harmless 4–20 page fixture with exact markers, rotation, transparency, selectable text, edge targets, and an image-only OCR page.
Dataset · 12 compression runs
Light vs Balanced vs Maximum compression benchmark
Compare output size, page and text retention, raster dimensions, and first-page fidelity for text, scan, photo, and slide-deck PDFs.
Compatibility lab · five HEIC fixtures
What survives HEIC-to-PDF conversion?
See measured results for EXIF orientation, pixel dimensions, transparency, Display P3 color, metadata removal, recompression, and browser network privacy.
Dataset · 8 fixtures · 10 OCR runs
OCR accuracy challenge: DPI, skew, rotation, noise, and language
Download image-only PDFs, Unicode ground truth, extracted outputs, character and word error rates, checksums, and settings guidance.
Dataset · 5 fixtures · 10 outcomes
PDF and Word fidelity: editable content vs visual layout
Compare searchable, editable, layout-aware, and visual modes with downloadable DOCX/PDF fixtures, converted files, page renders, exact checks, and checksums.
Local checker · print preflight
PDF page size, dimensions, and page-box checker
Inspect every page in inches, millimeters, and points; detect A4, Letter, rotation, mixed dimensions, MediaBox, CropBox, TrimBox, BleedBox, and ArtBox; export CSV or JSON.
Local checker · font preflight
PDF font embedding, subset, and ToUnicode checker
Inspect FontFile streams, Type 1/TrueType/CID/Type 3 objects, subset prefixes, encodings, ToUnicode maps, referenced pages, and extraction health; export CSV or JSON.
Local checker · PNG print preflight
PNG DPI, print size, transparency, and metadata checker
Read real PNG chunks, pHYs density, pixels, alpha/tRNS, profile signals, text metadata, CRCs, and exact A4/Letter PDF placement; export CSV or JSON.
Local checker · sharing preflight
PDF metadata and active-content privacy inspector
Inspect Document Info, XMP, attachments, form fields, external links, JavaScript events, launch actions, and parsed PDF object signals; export CSV or JSON.
Local checker · redaction preflight
PDF redaction checker: find text under black boxes
Compare positioned text with rendered dark regions, detect pending Redact annotations, search known sensitive phrases, and export the local evidence.
Local checker · image preflight
PDF DPI checker: effective image PPI and resolution
Measure every supported raster image placement at its actual physical size, compare with an adjustable target, and verify the math with a Poppler-tested fixture.
Data catalog · 13 open datasets
Open PDF research data catalog
Browse every benchmark and deterministic fixture in one catalog, with direct downloads, measured variables, methods, licenses, and machine-readable JSON.
Why the underlying evidence matters
PDF software can create a file that opens while still losing pages, changing page order, clipping an image, flattening transparency incorrectly, or damaging reading order. A reproducible test records the requested operation and inspects the output with an independent parser or renderer.
The same approach makes reference pages more useful. A DPI chart should expose its formula and rounding rule. A benchmark should name the sample file and validation method. A local-processing statement should describe the network test that would disprove it.
Reuse and citation
You may quote the factual measurements with a link to the relevant research page. The open data catalog provides direct dataset downloads, methods, and interpretation limits. Results describe the tested version and fixture; they are not a guarantee about every possible document or browser.