Open source on GitHub

This shop looks fine. A blind customer cannot use it.

On the left is an ordinary product page. On the right is what a screen reader actually announces to someone who cannot see it. Nothing has been exaggerated, and every fault is invisible to you.

What you seeRendered page
What a screen reader saysreading
    Runs axe-core in your own browser. Nothing is sent anywhere.

    Why the rule engine only found some of it

    What you just ran is a rule engine. It is genuinely good, and it is the first of three passes. But of the 55 success criteria the European Accessibility Act requires, here is how much of the job a rule engine can do alone. Each square is one criterion.

    Decided mechanically, 4
    Partly, 29
    Not at all, 22

    Point at a square. Each one is a success criterion you are legally required to meet.

    The difference, on one image

    Take the third product on that page. Its markup is <img alt="photo">.

    What a scanner concludes

    The image has an alt attribute. It is not empty. The rule is satisfied, and the scan moves on to the next element.

    1.1.1  pass

    What is actually true

    A customer who cannot see the screen is told the word “photo”. They do not learn that it is a 750ml trail flask, or that it costs £28. They cannot decide whether to buy it.

    1.1.1  fail, alternative text names the medium, not the content

    That judgement is most of an accessibility audit, and it is why a clean scan is not conformance. Our second pass is a language model reading the rendered page and the accessibility tree, looking for precisely this.

    And why you can believe the second pass

    Language models invent things. Telling a business it is breaking a law it is not breaking is worse than saying nothing. So no claim from the model reaches your report on the model’s authority. Each one has to survive five checks against your own captured page.

    • The success criterion it cites exists in WCAG 2.2 A or AA.
    • Its selector resolves to exactly one element we really captured.
    • The markup it quotes genuinely appears in that element.
    • It does not overturn a criterion the rule engine fully decided and found clean.
    • It does not restate a defect the rule engine already reported.

    Anything that fails is discarded and printed in your report as rejected, so you can see what it got wrong. A report that hides its mistakes is indistinguishable from one that never made any.

    Measured on a corpus written after the system was tuned, and run once: it found 5 of 7 defects the rule engine structurally cannot catch, with zero false positives on six deliberately correct elements. On the corpus we tuned against it scores 88%. We quote the 71%, because a score from the pages you tuned on is a score you gave yourself. The method and the known weaknesses are published.

    Run it on your own site, free

    The engine is open source. This is the same software that produced everything above, and there is no account, key or signup.

    $ npx playwright install chromium
    $ npx wcag-evidence https://your-site.example --report report.html

    Read a full report it produced, unedited, including the claims it rejected.

    What we sell is the part software cannot do

    The tool above will never tell you that you are compliant, because it cannot know. Twenty-two criteria need a person to sit down with a keyboard and a screen reader and actually use your site. That is the work, and it is what you are buying.

    Assessment

    €290
    One-off, up to 5 pages
    • Manual testing across the 22 criteria no software can settle: keyboard-only operation, screen reader, 400% zoom, error handling
    • The automated passes across your highest-risk pages
    • A written record of what was tested and what was found
    • Your accessibility statement, drafted for publication
    Ask for this

    Assessment and remediation

    €890
    One-off, up to 25 pages
    • Everything in Assessment, across your full purchase journey
    • A remediation plan ordered by legal exposure, so you fix what matters first
    • A call with your developers to walk through every finding
    • A re-audit after your fixes, so you hold dated evidence that you acted
    Ask for this

    Neither certifies conformance, because no honest audit can on its own. You will be told plainly what was established, what was not, and what remains.

    You may not need this at all

    The European Accessibility Act exempts microenterprises from its service requirements. If you have fewer than 10 staff and under €2 million in turnover or balance sheet total, you are very likely exempt, and you should not be paying anybody for this.

    Both must holdUnder 10 staff and under €2m. Exceeding either threshold removes the exemption.
    Services onlyIt does not extend to covered products.
    Groups count togetherParent, subsidiary and linked enterprise figures are aggregated.
    No grace periodThe exemption ends the moment you cross a threshold.

    If that describes you, use the free tool and keep your money.

    Ask for an audit

    Tell us the site, roughly how many pages matter, and any deadline you are working to. You will get a fixed quote and a start date, or an honest answer that you do not need this.

    Open an enquiry on GitHub

    A direct contact address is being set up. Until then GitHub is the reliable route.