PrivConvert public tests — September 6, 2026

These files are generated first-party test data. No customer files or personal
information are included. You may reuse these samples to repeat the tests.

Results were produced by selecting these inputs in the live privconvert.app
tool UI (revision 8f7181b), running the tools in Chrome 150 on macOS 15.3,
and downloading the outputs. Native FFmpeg generated INPUT clips only.

PDF: pc-text-source.pdf and pc-illustrated-source.pdf each have two pages,
selectable text, and one form field. Run Compress PDF in lossless, images,
and balanced modes; stripMetadata=false and onlyIfSmaller=false. The separate
pc-text-default.pdf uses balanced with onlyIfSmaller=true. The candidate in
that case was larger, so the original was returned unchanged.

Animation: use each pc-*-source.mp4 with both Video to GIF and Video to WebP.
Start=0 seconds, duration=6 seconds, width=480, FPS=10, WebP quality=80.
These are tool-output comparisons, NOT equal-perceptual-quality comparisons.
Encoded frame counts may differ because repeated frames can be combined.

results.json contains measured output bytes, SHA-256, PDF text/field counts,
and animation dimensions, durations, and encoded frame counts. qpdf and PDF
writers may change IDs or timestamps between runs, so new output hashes do
not necessarily match. The published hashes identify these specific files.

sample-manifest.json contains the input settings, generation commands, sizes,
and hashes. To regenerate input fixtures, install Node.js, pdf-lib, and an
FFmpeg build with libx264 and the lavfi filters used in the script. Run
generate-public-samples.mjs in a new working directory. It creates
public/evidence/2026-09-06 beneath that directory. FFmpeg or font/library
version differences may change exact bytes. Prefer the published input
files when comparing against this measured baseline.

analyze-public-samples.py requires Python with pypdf and Pillow. It inspects
existing downloaded results; it does not perform any browser conversion.
Place it in a scripts/ folder alongside the public/ folder before running.

network-observation.json summarizes one observed PDF run. It has explicitly
limited coverage: page-target requests, eight available POST bodies, one
unavailable POST body, and no independent worker-internal network capture.
It is NOT certification or evidence for all tools. No raw request headers,
cookies, tokens, or user records are published.

Guides:
https://privconvert.app/guides/pdf-compression-searchable-text
https://privconvert.app/guides/animated-webp-vs-gif
https://privconvert.app/privacy-proof

Corrections and questions: kevin.pyx@proton.me