Accessibility statement

Conformance status

Partially conformant with WCAG 2.2 level AA. Partially conformant means most of the standard is met but some parts are not. The gaps are listed below rather than left for you to discover.

What has been measured

The figures below were taken on 15 August 2026 against this build, by measuring the rendered document rather than reading the source. Contrast was computed against the background actually painted behind each element — sampled from a screenshot with the text hidden — because reading the CSS gives the wrong answer wherever a gradient is involved, and this site has one behind its largest heading.

Contrast

551 text nodes across the eleven public pages, with no exceptions. Every one meets the 4.5:1 minimum for body text or 3:1 for large text. The last shortfall was closed on 15 August 2026: the gold badges — “Expert-annotated”, “Teaching set” and the like — sat at 4.35:1 against their tinted background. Rather than bend an existing brand colour, the ramp gained a darker step, which the badge now uses at 6.3:1.

Layout

Zero horizontal overflow at 390px, 768px and 1280px, on all eleven pages. This was not always true: until 13 August 2026 the navigation bar never collapsed on a phone and the homepage overflowed a 390px viewport by 210px. The cause was a hydration mismatch — the bar decided its layout during the first client render, React declined to patch the resulting attribute difference, and the state that would have corrected it never changed. It is fixed and measured.

Keyboard and focus

• A skip link is the first tab stop on every public page, including the sign-in and registration screens, which did not have one until 13 August 2026, and the teaching-link page, which did not have one until 15 August 2026. The expired-link screen is the one exception, and deliberately so: it has no repeated block to bypass.
• Focus is visible on every interactive element, via :focus-visible and the design system’s focus ring.
• The role pills on the registration form are real radio inputs, so they are reachable and operable with the arrow keys; the visible ring follows the hidden input.
• Sign-in and registration are pages, not dialogs. No page on this site opens a modal, so there is no focus trap to get wrong. The design system carries a Dialog primitive that manages its own focus, but nothing currently uses it.

Targets

Every link and control measures at least 24×24px — the WCAG 2.5.8 AA minimum — at 390px, 768px and 1280px. Primary navigation and header actions are at least 44px tall. Three sets fell short and were corrected on 13 August 2026: the footer navigation links at 21px, the breadcrumb at 15px, and the links out of the sign-in and registration cards at 18px. Thirty-one links sit inside a sentence and are left at their natural height, which the standard permits.

Forms

The contact form marks each failed field with aria-invalid, points it at its own message with aria-describedby, and — as of 15 August 2026 — opens a summary of everything that failed, moves focus to it so it is announced, and leaves the next tab stop inside the form. Before that a screen reader user who pressed Send was told nothing at all: the button stayed, the page did not move, and the messages were scattered down a form they had no reason to re-read.

Structure and semantics

• lang is declared, and every page has exactly one h1 and a main landmark — verified on all eleven.
• Pipeline steps are an ordered list; the case listing is a list; the hero showcase is a labelled image group rather than three unlabelled images.
• Icons that sit beside their own text label are hidden from assistive technology instead of being read twice.
• Icons are drawn, not typed. The filter chip’s remove control was a × character until 13 August 2026 and is now an icon with an accessible name that says what it removes.

Motion

prefers-reduced-motion is honoured, but not by freezing everything. Reveals and counters resolve instantly. The hero showcase keeps advancing and keeps its cross-fade, because that preference is about movement and a fade has none; what it drops is the scale on the incoming slide. Auto-advancing content owes the visitor a stop regardless, so the pause control is available in every motion mode. If the reveal observer never fires, a failsafe shows the content anyway — a counter stuck at its start value would otherwise have rendered the words “0 annotated cases”.

Known limitations

The signed-in repository is unassessed

The sixteen pages behind sign-in — case search, the case page, the contributor editor, the DICOM viewer, the pipeline and its run screens, the admin queues — have had no contrast measurement, no target-size measurement, no keyboard pass and no structural check. They are built on the same design system and the same components as the pages that were measured, which is a reason to expect them to be better than nothing, not a reason to claim a result. This is the largest gap on this page.

Button sizes

The design system's small and medium buttons are 32px and 40px tall. Both clear the 24×24 minimum in WCAG 2.5.8 but fall under the 44px target guideline. Raising the base height would shift spacing across every screen, so it is being treated as a design decision rather than a defect.

Mobile navigation

Below 800px the navigation bar stacks into three centred rows, about 170px tall. It is no longer sticky there, so it scrolls away rather than occupying a fifth of the viewport, but the correct answer is a disclosure menu that has not been designed yet.

The DICOM viewer is third-party

The image viewer is OHIF, an external open-source application served from this origin. Its accessibility is not ours to state and has not been assessed here. A viewer is also, by its nature, a visual instrument; we do not currently offer a non-visual route to the same information.

How this was assessed — and what was not

Assessment was done by measuring the rendered document across the eleven public pages: contrast ratios against the background actually painted behind each element, horizontal overflow at 390, 768 and 1280px, computed target sizes at each of those widths, document structure, and the motion behaviour under both motion preferences. Browser console errors were captured at the same time and there were none.

Feedback

If you hit a barrier on this platform, tell us what you were trying to do, what got in the way, and what you were using — browser, operating system, and any assistive technology. That last part is what makes a report actionable.

Write to us through the contact page, choosing the technical topic. We aim to respond within five working days. If a fix will take longer, we will say what the workaround is in the meantime.

Standard applied

Web Content Accessibility Guidelines (WCAG) 2.2, level AA. Where this statement refers to a specific criterion — 1.4.3 contrast, 2.1.2 no keyboard trap, 2.4.1 bypass blocks, 2.4.3 focus order, 2.5.8 target size, 2.3.3 animation from interactions — it means that criterion specifically.