ToolkitPoint research
Original browser experiments with published methods, limitations, samples, and downloadable results.
What the research covers
The image format benchmark encodes the same three 800 × 500 synthetic scenes (photo-like, flat graphic and simulated document scan) in Chrome as PNG, at three JPEG quality levels and at three WebP quality levels. That gives 21 outputs, each with its file size and pixel error, and the CSV, JSON and script to repeat the run. Its numbers describe that browser encoder and those inputs, not every image.
The PDF compression benchmark evaluates five synthetic document topologies (multi-page text, archival document scans, hybrid diagram reports, interactive AcroForms, and widescreen presentation slides) across four client-side optimization modes, measuring exact byte reductions and verifying font text selectability with Mozilla PDF.js. All 20 test results, raw datasets, and reproduction scripts are published.
The privacy matrix classifies all 304 tools by what happens to the text or file you give them: kept in the browser, processed locally after a model download, handled by the browser's speech service, or one of the modes used by tools that need a network lookup. It is downloadable as CSV and JSON, so the classification can be checked instead of taken on trust.
PNG vs JPEG vs WebP
Compare 21 browser-generated outputs across three reproducible scenes, with file sizes, PSNR, samples, and raw CSV/JSON data. Headline findings: at representative settings, WebP 70 was 55% smaller than JPEG 75 on the photo-like scene (11.9 KB vs 26.8 KB at 38.11 dB PSNR), 61% smaller on the flat graphic (8.7 KB vs 22.1 KB), and 75% smaller on the document scan (5.1 KB vs 20.9 KB), while lossless PNG remained the largest file in all three scenes (93.8 KB to 366.4 KB).
BenchmarkPDF Compression: Size & Fidelity
Compare 20 client-side test runs across five synthetic document profiles. Key findings: Screen & Email compression reduced scanned documents by 66.9% and presentations by 67.0%, while Lossless Object Streams cut pure text documents by 37.0% and interactive forms by 36.4%—all with 100% verified text selectability.
Privacy datasetWhich tools send data away?
Inspect all 304 tools by processing mode, transmitted value, external dependency, and current deployment status. The dataset separates pure in-browser processing from tools that first download a local AI model, use the browser speech service, query a public data API, or run a server-backed domain or URL check.