Compliance Checker
We run axe and a vision model together against a page, then rank every barrier it finds by severity from Critical down to Minor.
Each finding arrives with the WCAG or Section 508 criterion behind it, a before-and-after of the markup, and a fix we can mark as worked, and the whole report sits on a compliance knowledge base we can question directly. The grounding model runs on local hardware, so nothing we scan ever leaves the bench.
How this works
A rules engine measures your page against WCAG. Those results are arithmetic. A contrast ratio either meets 4.5:1 or it does not, and you get the same numbers every run.
A vision model then looks at the rendered page three times, at three sensitivities. The tight pass reports only what it is sure about, and the looser passes catch what a second look would find.
Why the visual findings change
Review the same page twice and the visual findings can differ. That is expected, and it is why we show how often each one appeared.
The rules measure what is on the page. The vision model judges it, and a judgment about whether tap targets sit too close together lives near a threshold. Ask two experienced reviewers to analyze the same page and they will both flag the obvious problems, and each will catch something the other misses. It is the patterns that overlap that tell you more than any single pass.
- Found in 3 of 3 passes · consistent across every sensitivity
- Found in 2 of 3 · showed up in most passes
- Found in 1 of 3 · showed up once
What we confirmed, and what we noticed
Findings we verified a second way are marked confirmed. The rest are observations nothing automated can verify, like text baked into an image, and those need your eyes. Every check runs on our own hardware, so nothing you scan leaves the bench.