Open evidence for PDF software
Open PDF research data catalog
Eleven reusable datasets collect the site’s original benchmark results, deterministic fixtures, expected-result manifests, checksums, and independent validator output. Every cataloged dataset is free to inspect and licensed CC BY 4.0.
Use the Schema.org JSON-LD catalog for structured metadata, llms.txt for the concise site map, or llms-full.txt for citation-ready facts and interpretation limits. The canonical HTML dataset pages remain authoritative.
Open JSON-LD catalogOpen full AI evidence fileRead validation method
Browse all downloadable datasets
| Dataset | Question answered | Primary downloads | Independent evidence |
|---|---|---|---|
| Browser PDF tools accuracy | Do 22 production browser workflows produce the requested result? | Results CSV | Parsing, rendering, ZIP checks, marker comparison, SHA-256 |
| PDF compression presets | How do Light, Balanced, and Maximum affect size and fidelity by document type? | Results CSV | Poppler text/images and 120-DPI render comparison |
| HEIC-to-PDF compatibility | What happens to orientation, alpha, dimensions, metadata, color, and privacy? | Results CSV · Fixture manifest | libheif, Poppler, render metrics, Chrome network trace |
| Browser OCR accuracy | How do DPI, noise, skew, rotation, contrast, and language affect OCR? | Results CSV · Fixture ZIP | Unicode-aware character and word error scoring |
| PDF/Word fidelity | Which conversion modes preserve editable content versus visual layout? | Results CSV · Detailed checks · Fixture ZIP | PDF objects, DOCX XML, LibreOffice rendering, exact markers |
| PDF page boxes | What are the exact dimensions, rotations, and five effective page boxes? | Fixture PDF · Manifest JSON | Independent pypdf expected results |
| PDF fonts and ToUnicode | Which fonts are embedded, subset, encoded, mapped, or missing? | Fixture PDF · Manifest JSON · Poppler evidence | Poppler pdffonts |
| PNG print readiness | What do pixels, pHYs, transparency, profiles, metadata, CRCs, and placement mean? | Manifest JSON · ImageMagick evidence | ImageMagick identify |
| PDF privacy and active content | Does a PDF contain metadata, XMP, attachments, forms, links, JavaScript, or actions? | Positive fixture · Negative fixture · Manifest JSON | Poppler pdfinfo and pdfdetach |
| PDF redaction failures | Is text truly removed, merely covered by black, or waiting in a Redact annotation? | Unsafe fixture · Safe fixture · Pending annotation | Poppler text/raster output and ImageMagick |
| PDF effective image PPI | What is each raster image’s effective PPI at its actual physical placement? | Fixture PDF · Manifest JSON · Poppler evidence | Poppler pdfimages -list |
How to cite or reuse a result
- Name the dataset and tested case.
Do not cite a broad claim such as “all PDF compression preserves quality” when the evidence covers a named fixture and preset.
- Link the canonical dataset page.
The page explains scope, generation, validator versions, assumptions, and limitations that a raw CSV cannot carry alone.
- Preserve the version or checksum.
Use dataset query versions and recorded SHA-256 hashes when exact reproduction matters.
- Attribute under CC BY 4.0.
A suitable attribution is: “GoPDFConverter [dataset name], Michael Joiner, 2026,” linked to the canonical page.
What this catalog does not prove
A fixture demonstrates the tested case, browser path, and release—not every possible PDF, PNG, HEIC, DOCX, browser, printer, or competing product. A “pass” means the published expected checks passed. It is not a compliance certificate, security audit, or universal guarantee. Each dataset page publishes its own limitations and should be read before generalizing a result.
The catalog contains synthetic or non-sensitive fixtures. It does not collect uploaded user documents. Tools analyze selected files locally in browser memory; public dataset downloads are ordinary static web assets.
Machine-readable access
Schema.org DataCatalog JSON-LD
Eleven Dataset records with descriptions, methods, variables, keywords, licenses, and direct distributions.
Concise LLM discovery file
Canonical links and one-line descriptions for tools, guides, research, ownership, and interpretation limits.
Full AI evidence file
Citation-ready factual summaries, exact measured results, source links, and explicit limits in plain Markdown.
Atom research feed
Chronological discovery for newly published datasets, guides, and significant updates.