The count
axe-core 4.10.3 ships 69 rules in its WCAG A and AA tag sets. Those rules carry WCAG tags for 23 of the 55 success criteria that WCAG 2.2 defines at Level A and AA. The remaining 32 have no rule tagged to them at all.
That is not a criticism of axe-core, which is the best engine of its kind and is careful about what it claims. It is the shape of the problem. The criteria a machine can settle ask whether something exists: does this image have alternative text, does this control have a name, does this text meet a contrast ratio. The criteria below ask whether something makes sense, and meaning is not in the markup.
Pair this with the other half of the picture: in our scan of 410 high-traffic home pages, 74% failed at least one check a machine can decide, and 81% also returned checks it would not decide. Both numbers point the same way. The automatic part is cheap. The judgement is the work.
The 32, and what each one needs from a person
| Criterion | Name | Level | What a person has to do |
|---|---|---|---|
| 1.2.3 | Audio Description or Media Alternative (Prerecorded) | A | Watch the video. Decide whether the audio description or the text alternative actually conveys what is on screen. |
| 1.2.4 | Captions (Live) | AA | Attend or sample the live stream and judge whether the captions keep up and are accurate. |
| 1.2.5 | Audio Description (Prerecorded) | AA | Watch with the audio description on and check it covers what matters visually. |
| 1.3.2 | Meaningful Sequence | A | Turn off CSS, or read the page with a screen reader, and check the reading order still makes sense. |
| 1.3.3 | Sensory Characteristics | A | Read every instruction and look for ones that rely on shape, size, position, sound or colour alone: “click the round button on the right”. |
| 1.4.5 | Images of Text | AA | Look for text baked into images. A tool cannot tell a logo from a paragraph in a picture. |
| 1.4.10 | Reflow | AA | Set the window to 320 CSS pixels wide and check nothing needs two-dimensional scrolling and nothing is lost. |
| 1.4.11 | Non-text Contrast | AA | Check that the parts of a control that carry meaning — its border, its icon, its focus ring — have 3:1 contrast. Which parts carry meaning is a judgement. |
| 1.4.13 | Content on Hover or Focus | AA | Hover and focus every tooltip and popover. Check it can be dismissed, hovered over, and stays until dismissed. |
| 2.1.2 | No Keyboard Trap | A | Tab through the whole page, including every widget and modal, and check you can always get out again. |
| 2.1.4 | Character Key Shortcuts | A | Look for single-character shortcuts and check they can be turned off, remapped, or only fire on focus. |
| 2.3.1 | Three Flashes or Below Threshold | A | Watch any animation or video for flashing. Measuring it needs a person and, for anything borderline, a tool like PEAT. |
| 2.4.3 | Focus Order | A | Tab through the page and judge whether the order preserves meaning and operability. Only a person knows what the intended meaning was. |
| 2.4.5 | Multiple Ways | AA | Check there is more than one way to reach each page: navigation, search, a sitemap. |
| 2.4.6 | Headings and Labels | AA | Read the headings and labels and judge whether they describe what follows. A tool sees that a heading exists, not whether it is honest. |
| 2.4.7 | Focus Visible | AA | Tab through and look. Check the focus indicator is visible on every control, against every background it lands on. |
| 2.4.11 | Focus Not Obscured (Minimum) | AA | Tab through and check nothing sticky — a header, a cookie bar, a chat bubble — hides the focused control. |
| 2.5.1 | Pointer Gestures | A | Find every gesture that needs a path or multiple fingers and check there is a single-pointer alternative. |
| 2.5.2 | Pointer Cancellation | A | Press and hold on controls and check the action fires on release, and can be aborted by moving away. |
| 2.5.4 | Motion Actuation | A | Shake, tilt or move the device and check any motion-triggered action has a conventional alternative and can be switched off. |
| 2.5.7 | Dragging Movements | AA | Find every drag interaction and check there is a way to do the same thing with a single tap or click. |
| 3.2.1 | On Focus | A | Tab into every control and check nothing changes context on focus alone. |
| 3.2.2 | On Input | A | Change every input and check nothing submits, navigates or reorders the page without being asked. |
| 3.2.3 | Consistent Navigation | AA | Compare navigation across pages and check the order is consistent. That needs more than one page, which a single-page scan never sees. |
| 3.2.4 | Consistent Identification | AA | Compare icons and controls that do the same job across the site and check they are named the same way. |
| 3.2.6 | Consistent Help | A | Check that help — contact details, a chat, a help link — appears in the same relative place on every page that has it. |
| 3.3.1 | Error Identification | A | Trigger every error and check the message says which field is wrong, in text. |
| 3.3.3 | Error Suggestion | AA | Trigger every error and judge whether the suggestion is useful and correct. |
| 3.3.4 | Error Prevention (Legal, Financial, Data) | AA | Walk through anything legal, financial or destructive and check it is reversible, checked, or confirmed. |
| 3.3.7 | Redundant Entry | A | Complete a multi-step process and check it does not ask again for information you already gave. |
| 3.3.8 | Accessible Authentication (Minimum) | AA | Try to sign in without a cognitive function test, or check the alternative or the object-recognition exception applies. Password managers must be able to fill it. |
| 4.1.3 | Status Messages | AA | Trigger status messages — “item added”, “3 results”, “saving” — and check a screen reader announces them without moving focus. |
The pattern worth learning
Read the right-hand column and the same four verbs keep appearing: tab through, trigger, compare, judge. Three of those are mechanical. Only the last one needs a person.
- Tab through covers focus order, keyboard traps, focus visibility and focus obscured. Somebody or something has to actually walk the page with a keyboard.
- Trigger covers every error-handling criterion. You cannot check an error message that never fired, which is why error states are the most commonly missed part of an audit.
- Compare covers consistent navigation, consistent identification and consistent help, and it needs more than one page. A single-page scan is structurally incapable of it.
- Judge is the residue: is this heading honest, is this suggestion useful, does this order preserve meaning. That is the part no tool takes off you, and the part worth your attention.
An agent can do the first three and hand you the fourth with the evidence attached. That is the whole argument for agentic testing, and it is why our reports carry four states rather than two: verified, issue found, needs input, not covered. A criterion nobody checked is not covered, and saying so is more useful than a green tick that meant "we did not look".
How to use this list
- Do not treat it as your whole manual audit. It is the floor: the criteria you definitely cannot delegate to a scanner. Several of the 23 a tool does check still need a person to confirm the result.
- Start with the four you will fail. In most audits the first failures are 2.4.7 focus visible, 2.4.3 focus order, 3.3.1 error identification and 1.4.11 non-text contrast. They cost little to fix and they are invisible to every scanner.
- Run the cross-page ones once per release, not per page. Consistent navigation, consistent identification and consistent help only mean anything across a set.
- Write down what you did not check. That single habit is the difference between an audit and a reassurance. Our website QA checklist marks which checks a tool can decide and which need a person, and European Accessibility Act testing covers what the law now expects of that record.
- Work through all 55, not only the 32. Our WCAG 2.2 quick check is a free, in-browser checklist covering every Level A and AA criterion, grouped by principle, with this page's 23-of-55 split marked against each one as you go.
Method
- The tool side. axe-core 4.10.3, loaded in a real browser, queried with
axe.getRules(["wcag2a", "wcag2aa", "wcag21a", "wcag21aa", "wcag22aa"]). Each rule carries tags of the formwcag111, which map to success criterion numbers. 69 rules, carrying tags for 24 distinct criteria, one of which (2.1.3 Keyboard, No Exception) is Level AAA and so falls outside this comparison, leaving 23 at A and AA. - The standard side. The success criteria listed in the W3C WCAG 2.2 Recommendation, read from the Recommendation itself: 31 at Level A and 24 at Level AA, 55 in total. Level AAA is out of scope, as it is for almost every legal requirement.
- The difference is the 32 above. "No rule tagged to it" is the precise claim. A rule may happen to catch something related to a criterion it is not tagged with, and axe's incomplete results flag some of this territory for review rather than ignoring it.
- Other engines differ at the edges and none of them close this gap, because the gap is not a tooling shortfall. It is what the criteria ask.
Sources, opened and checked 21 September 2026: W3C, Web Content Accessibility Guidelines (WCAG) 2.2; W3C WAI, Selecting Web Accessibility Evaluation Tools; Deque, axe-core 4.10.3; ShipperAG, what automated accessibility testing catches, and what it misses (20 September 2026).