The method
Published in full, checklist included. If you want to run it yourself, you can. What you buy is someone running it who has run it before and knows what to look at when something does not fit the pattern.
37
checks
19
need a person
4
WCAG principles
WCAG 2.2 Level AA
Structure
The audit follows the four WCAG principles rather than a list Uxerfy invented. The deliverable has to map onto the standard your legal counsel, a procurement questionnaire or an external auditor will ask about.
Whether the information on the page reaches someone who cannot see it, cannot see it well, or cannot distinguish the colours it is drawn in. Content that exists only as a picture, a colour or a contrast too faint to resolve is content that some of your customers do not have.
Whether the page can be used without a mouse. Keyboard access is the principle most often broken by modern component libraries, and the one automated tooling is worst at detecting — a custom dropdown that traps focus passes every scanner and fails every keyboard user.
Whether the page behaves predictably and tells people how to recover. Most of the cost here lands in forms: an error that is announced only in red, a field whose label disappears once it is filled, a session that expires without warning during a checkout.
Whether assistive technology can tell what each control is, what state it is in, and when something on the page has changed. This is where custom components fail: a div styled as a button is a button to everyone except the person who cannot see it.
Evidence
18/37
Detected by tooling against the page source. Reproducible, fast, and blind to anything that depends on interaction, sequence or what a screen reader actually announces.
19/37
Found by a person operating the page with a keyboard and a screen reader. Slower, not reproducible by a script, and the only way most serious barriers are ever found.
Checklist
Each check states the question, what it costs when it fails, how to test it, and the success criterion it maps to.
The cheapest findings in an audit live here, and so do some of the most expensive. Alternative text is a fifteen-minute fix; a design system whose entire palette sits at 3.9:1 is a quarter of work.
The commonest failureThe commonest failure is alternative text that describes the file rather than the function — "icon-arrow-right.svg", or a product photo whose alt is the SKU. The attribute is present, the scanner passes, and the information is still missing.
This is the half of the audit that no scanner performs, and the half where the findings that end in complaints are found. Everything here is discovered by a person with their hands on a keyboard.
The commonest failureThe commonest failure is a custom component — a dropdown, a modal, a carousel — that works perfectly with a mouse and is unreachable, or inescapable, with a keyboard. It ships because it was only ever tested with a mouse.
Forms are where this principle is won or lost, and forms are where the money is. A checkout that announces its errors properly converts better for everyone, not only for the people who need it to.
The commonest failureThe commonest failure is an error message that appears visually, in red, next to the field — and is never announced, never associated with the input, and never focused. The page looks like it is telling you; it is telling only some people.
Short list, high stakes. Almost everything here concerns custom components, and almost every failure is introduced by a framework or a component library rather than written deliberately.
The commonest failureThe commonest failure is ARIA applied as a decoration — aria-label on a div that is not focusable, role="button" without a keydown handler, aria-expanded that never changes. Incorrect ARIA is worse than none, because it makes a confident claim that turns out to be false.
Severity
Severity answers what a barrier costs the person who hits it, never how hard it is to fix. Mixing the two is how a one-line CSS change ends up above a checkout that cannot be completed by keyboard.
Someone using assistive technology cannot complete the task at all. A checkout that cannot be reached by keyboard, a form whose errors are never announced, a dialog that traps focus. This is the category that ends in a complaint.
The task can be completed, but only with difficulty, guesswork or luck. An unlabelled field that can be inferred from position, a focus indicator that disappears against one background, a heading structure that makes the page hard to navigate by landmark.
A real barrier for some people in some conditions. Contrast that passes on desktop and fails on a phone in daylight, a link whose purpose needs its surrounding sentence, a target smaller than 24px.
Worth fixing while the file is open. Redundant alternative text, a decorative image that is not hidden, a language attribute missing from a quoted phrase.
Score
Each principle starts at 25 points and loses points for each failing check, weighted by severity. The total is the sum. The arithmetic is published because a number the client cannot reproduce is a number they are being asked to take on faith, which is the opposite of what an audit is for.
A check that fails on many elements is one finding, not many. Past 10 instances it raises the severity by one level, but it never multiplies the deduction: a score that collapses on a single repeated failure stops describing the page.
This score is an internal Uxerfy assessment indicator based on the checks included in this review. It is not a legal certification or a statement of WCAG conformance.
Conformance. This is an internal Uxerfy assessment indicator based on the checks in this scan, not a statement of WCAG conformance and not a legal certification.
Everything WCAG covers. Around a third of WCAG 2.2 success criteria can be decided automatically at all; the rest require a person, and some require testing with disabled users.
Whether people with disabilities succeed on your site. That needs usability testing with those users, which is a different engagement from an audit.
Content quality — captions that exist but are wrong, alternative text that is present but inaccurate, transcripts that are auto-generated and unedited.
Anything behind a login, a paywall or a region block that Uxerfy was not given a way through.
Native mobile applications, PDFs and documents, unless explicitly included in the scope.
The method is public. The difference is who runs it on your site and what they find.
Sergio Gualda · Barcelona, Spain