The testers run on our own model, not a public chatbot

• Your pages stay with usSite text and reports are processed on servers we operate and never go to a public AI service.
• No developer filtersWe run an uncensored build of the model, with the refusals and guardrails of public AI services removed. A tester can say bluntly that a page looks like a scam or a price is too high.
• Accuracy we measure, not promisePeople in our pilot check each note on their own sites and mark it true or wrong. The share of true notes will appear here once enough are checked.

What the model sees, and what it does not

• It reads the text and the controls on the screen, the way a screen reader would, and decides what to do next as the persona.
• It does not judge the visual design: colours, spacing and images are outside what it reads.
• It can be wrong. Every note can be marked true or wrong, and the wrong ones never reach the fix file.