Evidence
How NexKit Tools Are Actually Tested
This page is the index for a simple rule: any specific factual claim about how a NexKit tool behaves — that it runs locally, that compression is content-dependent, that a limit is a real number — should be backed by a test you can trace, not just a description. This page holds the methodology and the measured-results summary; full write-ups live on the Guides blog.
Methodology
Each test here uses generated, copyright-safe fixture files run through a scripted browser session against the live production tool — not a staging environment, not a description of intended behavior. Every claim traces to an actual recorded run: file sizes, timing, and network requests captured during the test, not estimated afterward.
When a result doesn't reproduce consistently, we say so rather than quietly picking the more favorable number. Every test lists its own known limitations, including where the test fixtures may not represent real-world files.
Measured results
| Tool | Fixture type | Result | Full report |
|---|---|---|---|
| Compress PDF | Text-only, image-heavy, pre-optimized, and raw-image PDFs | 18 identical automated repeat trials produced no smaller output; one earlier manually observed 2% reduction was not reproducible; no additional file-upload network request was observed during the tested processing runs; the tool consistently declined to offer an equal-or-larger candidate. | Read the report |
| Merge PDF / PNG to JPG | Generated multi-page PDFs; generated PNG image | Zero network requests recorded during merge processing; the only network-layer event during image conversion was a browser-internal blob: reference, not a request to any server. Both tools reported real, verifiable output stats (merged page count matched input exactly; image conversion showed a genuine 51.9% size reduction). | Read the report |
| Add Page Numbers | 299/300/301-page PDFs; AES-256 password-protected PDF; deliberately truncated (damaged) PDF | The 300-page limit held exactly as stated: 299 and 300 pages accepted and processed correctly, 301 pages rejected with a specific on-screen error. Password-protected and damaged files were each rejected with a distinct, accurate error message. The 25MB size boundary was not tested in this round. | Read the report |
Known limitations
These tests use synthetically generated fixtures, not real user documents — results on real-world files with different internal structure may differ.
Each test describes one tool under one set of conditions (browser, date, production state at the time). Results are not assumed to generalize to other tools or other conditions unless separately tested.
Network-request observations describe what was recorded during the specific test runs listed here, not a general guarantee covering every use of a tool.