The target, what is tested, and what is not.

What is tested, and how far it goes.

• 25 public pages are rendered in a real browser and audited automatically on every run.

  A named list: the homepage, the four applications and their catalogue, how buying works, the questions pages, the demonstrations index, the Tools catalogue, pricing, Custom Agents, About, Contact, Careers, Security, Trust, Accessibility, Status, Privacy, Terms, Search, sign-in and sign-up. A page not on that list is not covered by this claim.

• A serious or critical finding fails the build.

  Nothing ships past one. Minor and moderate findings are printed with the page they came from and decided deliberately, rather than turning the suite red over a rule nobody has made a decision about.

• The rules checked are the ones a machine can decide: contrast, accessible names, roles, landmarks and form labels.

  That is roughly a third of what accessibility means and none of what a person operating the page would tell you. It is a floor.

• Keyboard operation is covered by its own tests.

  Focus stays visible, the mobile navigation traps focus, closes on Escape, and returns focus to the control that opened it.

• Reduced motion never changes a first render.

  A test asserts it, so somebody who asked for less motion does not get a broken page instead.

• Colour is never the only carrier of meaning.

  A status reads as a word before the dot beside it is reached, on every surface that has one.

What that scope leaves out.

An automated audit passing is not the same as a page being usable. These are the parts this page cannot vouch for.

• No screen reader testing is automated. Real assistive technology is used by hand and irregularly, and that is the largest gap on this page.
• No page behind sign-in is in the audited list. The twenty-five are public pages.
• No formal accessibility conformance report exists, and no third party has assessed this site.
• Touch targets are designed to be comfortable rather than merely compliant, which has not been measured systematically across every control.
• The applications sold here are audited separately and ship with their own results. An accessibility defect in one is something a buyer inherits knowingly rather than discovers later.

What checking has caught, with dates.

History rather than a score. These were found by rendering a page, and they are here because the kind of defect that gets through is more informative than a count of the ones that did not.

• A danger tone on a danger surface, at 3.92 to 1.

  Fixed · found 2026-09-14

  Below AA, when the token drawn for exactly that pairing already existed.

• The same mistake again, in a banner's informational tone.

  Fixed · found 2026-09-14

  A different surface, found the same way.

• The site's most-used secondary text colour, at 3.94 to 1.

  Fixed · found 2026-09-14

  On nearly every page. It is now 4.5 to 1 or better, which is why the body text here reads the way it does.

• A thumbnail beside the same name in text, reported as a missing text alternative.

  Withdrawn · found 2026-10-30

  Not a defect. An empty alt attribute is correct there: announcing the name a second time is noise, not information. The report was withdrawn and this is recorded so it is never 'fixed'.