Compression
Record input/output bytes, ratio, elapsed time, preset, and whether output parsing succeeds.
Measured result: not published
No-upload verification method
This procedure tests a specific tool, fixture, browser, and app version. It does not prove that the browser makes zero requests or permanently certify future releases.
Manual browser procedure
- 1
Choose a safe fixture
Download a synthetic PDF from the verification hub and record its SHA-256 digest.
- 2
Record network traffic
Open browser developer tools, select Network, preserve the log, and clear earlier requests.
- 3
Run one workflow
Select the fixture, complete the tool operation, and download the result.
- 4
Inspect requests
Search request bodies and content types for PDF, image, multipart, filename, or fixture-byte transmission.
- 5
Record boundaries
Save the app commit, browser version, tool, preset, result, and any ordinary resource requests.
What ordinary traffic may remain
Page HTML, JavaScript, workers, fonts, PDF.js, Tesseract, and qpdf WASM resources may load over the network. Consent-gated analytics can send bounded event metadata. The test is specifically about requests containing document bytes or user document data.
Automated repository check
Playwright workflow tests observe requests while safe fixtures are merged, protected, and processed by OCR. They fail when a request carries PDF, JPEG, PNG, or multipart content. Workflow tests also require non-empty downloaded outputs.
This is implementation evidence, not an independent security audit. Repeat the manual procedure after dependency, hosting, analytics, or workflow changes.
Benchmark protocols without invented results
OCR
Use a reviewed scanned fixture, known ground truth, character error rate, language, pages, and elapsed time.
Measured result: not published
Encryption
Record input/output bytes, elapsed time, qpdf preset, parse rejection without a password, and authorized decryption verification.
Measured result: not published