WCAG 2.2 Demo SuiteStep 4 — Audit the selected sample
Check every page in your sample against every criterion in scope. Record pass, fail, or not applicable — with evidence for each failure, because a finding nobody can reproduce will not get fixed.
4a. Check for the criteria in scope
For each page, work through the criteria for your target level. Use the testing sequence: automated scan, keyboard pass, zoom and reflow, text spacing, screen reader, error states, content review.
Three outcomes per criterion per page:
| Outcome | When | Record |
|---|---|---|
| Pass | The criterion is met on this page | How you checked it |
| Fail | At least one instance does not meet it | Location, evidence, who it affects, and the fix |
| Not applicable | There is nothing on the page the criterion applies to | Why — "no video on this page", not just "N/A" |
Do not invent a fourth outcome. "Partially passes" is not a thing: WCAG conformance is binary per criterion per page. If one image in twenty lacks alt text, 1.1.1 fails on that page.
4b. Record what you find
A finding that cannot be reproduced will be argued with and then dropped. Every failure needs six things:
- The criterion — number, title and level.
- The location — page URL and a CSS selector or unambiguous description.
- What happens — the observed behaviour, not your interpretation of it.
- How you found it — the tool, browser, assistive technology, and steps.
- Who it affects, and how badly — "a screen reader user cannot complete checkout" lands very differently from "missing label".
- A suggested fix — usually a technique from the W3C Understanding document.
Severity
WCAG has no severity scale — every failure is equally a failure for conformance. But a report that does not prioritise is unusable, so rank the findings separately from the conformance result:
| Severity | Meaning | Example |
|---|---|---|
| Blocker | A group of users cannot complete an essential task at all | A keyboard trap in the checkout modal |
| Serious | A task can be completed, but with substantial difficulty or risk of error | Errors signalled by colour alone |
| Moderate | Noticeable friction; workarounds exist | Vague link text in a list of articles |
| Minor | Small barrier, low frequency | A decorative image with unnecessary alt text |
4c. Check for accessibility support
Verify that the techniques the site relies on actually work in the baseline you defined. An ARIA pattern that works beautifully in one screen reader and is silent in another is a failure if both are in your baseline.
4d. Archive your evidence
Keep screenshots, screen reader transcripts, the scanner's raw output, and the exact URLs and dates. Sites change; six months later your report is the only record of what you saw.
Aggregate before you report
The same defect usually appears on many pages, because it lives in a shared template or component. Report it once, list the pages it affects, and fix it once. A register of 240 separate findings that are really 12 defects buries the signal and exhausts the team.