WCAG criteria Rampa judges
axe-core decides what markup can prove. For nine WCAG 2.1 success criteria, Rampa also asks a model what rules can’t answer: whether an alt says what the image shows, a field names its purpose with the right autocomplete, a picture of text could be real text, a title names the page, a link says where it goes, a heading or label describes its content, lang names the language of the text, and a field’s label is on screen. An answer reaches the report only when the text it quotes is on the page.
| Criterion | axe-core checks | Rampa judges | Measured on |
|---|---|---|---|
| WCAG 1.1.1 Non-text ContentLevel A | an image has a text alternative | the alternative says what the image shows, seen as it renders | ACT qt1vmo: F1 1.00, 16 cases |
| WCAG 1.3.5 Identify Input PurposeLevel AA | an autocomplete value is a valid token for its field | a field for the user’s own data names that purpose with the right token | Corrupted pairs: 14 of 16 told apart |
| WCAG 1.4.5 Images of TextLevel AA, runs on request | no rule: the question is about pixels | a picture that is mostly text, logos aside, could be real text | ACT 0va7u6: F1 0.75, 15 cases |
| WCAG 2.4.2 Page TitledLevel A | the page has a non-empty title | the title names this page’s topic or purpose | ACT c4a8a4: F1 1.00, 6 cases |
| WCAG 2.4.4 Link Purpose (In Context)Level A | a link has a name | the name, with the text around it and the page it leads to, says where the link goes | ACT 5effbb: F1 0.92, 18 cases; ACT fd3a94: F1 0.70, 24 cases |
| WCAG 2.4.6 Headings and LabelsLevel AA | no rule mapped to 2.4.6 | a heading describes the content under it; a label says what to enter | ACT b49b2e: F1 0.89, 12 cases; ACT cc0f0a: F1 0.91, 16 cases |
| WCAG 3.1.1 Language of PageLevel A | lang is present and valid | lang names the language most of the page is in | ACT ucwvc8: F1 0.67, 14 cases |
| WCAG 3.1.2 Language of PartsLevel AA | lang values are valid | each passage with a lang is in that language | ACT off6ek: F1 0.89, 13 cases |
| WCAG 3.3.2 Labels or InstructionsLevel A | a field has a name, and not two labels | the name is on screen, and an enforced format is explained | Corrupted pairs: 7 of 7 told apart |
Of the 50 WCAG 2.1 success criteria at levels A and AA, a run on the store example checks 12, judges 7 of them with evidence, and leaves 38 to people. Passing these checks does not make a page accessible.
What every judgment goes through
Collect
The page becomes a normalized accessibility snapshot: roles, names, languages, states, and every image as it was rendered.
Rules first
axe-core runs on the page. Whatever its rules can decide goes straight to the report, with no model involved.
Judge the residue
What rules can’t decide goes to a model: one prompt per WCAG criterion, the smallest context that answers it, and a typed answer back.
Verify the evidence
Every claim must cite a node that exists and text that is really in it. A claim that fails is dropped and counted, never shown.
Report the coverage
Findings come with a patch, and the report lists what nobody checked. It never calls a page accessible.