On 24 October I found out that my own responsive audit had been reporting a width it never reached. The extension asked Chrome for a 320 px window, Chrome stopped at roughly 500 px (that's the narrowest window I could get on Windows, and I haven't checked every platform), and the report still drew a green "320 ✓" next to a page that was never 320 px wide. The same check at 768 px was really measuring 755, because the window frame and the scrollbar took 13 px.
Nobody had complained, which bothered me more than the bug itself, because a layout that breaks only on narrow screens would have got a green tick too. Version 1.2.0 measures the real CSS width at every step, and the way it reaches 320 px comes from the WCAG Understanding document.
Why 320 px, and how zoom gets you there
WCAG 2.2 success criterion 1.4.10 Reflow asks that content can be presented without scrolling in two directions at a width equivalent to 320 CSS pixels. The note under it explains the number: 320 CSS pixels is what a 1280 px wide desktop window becomes at 400% zoom (Understanding 1.4.10). So the criterion is as much about people who zoom the page because they need larger text as about small phones.
That equivalence is also the workaround for Chrome's minimum window size. The audit keeps the narrowest window Chrome allows and sets the tab zoom to whatever turns it into 320 CSS px (about 1.6× from a 500 px window on my machine). It zooms only the tab being checked and restores the original zoom afterwards. There's a side effect you should know about: the page sees a higher device pixel ratio (about 1.95 at "320" on my laptop), so a site with responsive images may load larger files than a real phone would, and media queries based on resolution behave accordingly. The report says when a width was reached through zoom.
How to check a responsive layout automatically
Run an audit that sets each width in turn and reports horizontal scroll with the element that causes it, because the element is what you actually fix. In my experience it's usually something with a fixed width (a table, an embedded map, an image without max-width) or an unbreakable string such as a long URL or an email address in the footer. WCAG exempts content that needs a two-dimensional layout, such as data tables and maps, so the fix is usually to let the table scroll inside its own container rather than the whole page. Besides the scroll, the audit reports blocks that overflow their container, items that dropped out of their row onto a new line (in a flex row with wrapping or a row of inline-block or floated items), text clipped by overflow: hidden, overlapping elements and broken images. In UI Testing Inspector it runs at 320, 768, 1024 and 1440 px by default, and you can set your own widths.
You can do the same by hand in DevTools device mode, and for one width that's often quicker. The extension is worth it when you want all four widths in one report with the offending elements marked, or when you need to send the result to someone who never opens DevTools. For a before and after check of the same page, a saved baseline does more than device mode, and I describe that workflow in visual regression testing without CI.
Three WCAG 2.2 criteria you can check with geometry
Some accessibility criteria are about meaning and need a person, but a few are about boxes on the screen, which a browser extension can measure exactly. Two of the three below, 2.5.8 and 2.4.11, are new in 2.2. Criterion 2.5.8 Target Size (Minimum), level AA, asks for pointer targets of at least 24 by 24 CSS pixels, with an exception for smaller targets spaced so that 24 px circles centered on them don't intersect another target or another small target's circle (Understanding 2.5.8). The check reports small links and buttons only when that circle actually touches a neighbor, so a lone 20 px icon with enough space around it isn't flagged. Links inside a sentence are exempt too, and the check skips them.
Keyboard users run into a different problem on pages with sticky headers. Criterion 2.4.11 Focus Not Obscured (Minimum), also AA, says that a component receiving keyboard focus mustn't be entirely hidden by content the author created, and the Understanding document names sticky headers and footers as the typical cause (Understanding 2.4.11). The extension focuses each focusable element in turn (up to 500 on very large pages), the way a keyboard user would move through the page, and reports the ones that end up completely under a fixed bar. Partly covered elements pass, as the criterion allows, and semi-transparent bars don't count as hiding.
The last one has exact numbers. Criterion 1.4.12 Text Spacing asks that nothing is lost when a user sets line height to 1.5 times the font size, paragraph spacing to 2 times, letter spacing to 0.12 times and word spacing to 0.16 times (Understanding 1.4.12). The audit applies exactly those values and shows what gets clipped or starts to overlap. There's also a stress mode that makes every text 35% longer, which isn't a WCAG criterion but catches the buttons that break after translation into German or Russian.
What this doesn't cover
This isn't a full accessibility audit. Contrast, labels, heading structure and keyboard order need a tool such as axe or a manual review, and real devices still matter for touch behavior and Safari. I also couldn't find a ready-made automated rule for 2.4.11 in axe-core or the W3C ACT rules, so that check is my own reading of the Understanding document, and if you find a case where it's wrong, I'd like to hear about it.
Links: WCAG 2.2 · Understanding 1.4.10 Reflow · UI Testing Inspector FAQ · UI Testing Inspector in the Chrome Web Store
P.S. The 320 px bug turned up during a planned review of the audit code and not in a user report. I'd rather it had been caught earlier, but at least the fix relies on the same zoom equivalence W3C uses for testing Reflow.
Ana Do builds UI Testing Inspector and writes about checking layouts and measuring how reliable AI systems are. More writing and research at anaexplores.github.io.