SquishPixel Blog · Accessibility · Tools · Front-end

Automated scan caught 10 accessibility issues in a health app; manual testing found 7 more

According to DEV Community, an author audited OpenMRS O3, an open source electronic health record, against WCAG 2.2 level AA. WCAG is the standard guide for accessible web content. The audit used version 3.7.1 run locally from Docker images, and covered eight screens, including login, patient search, patient registration and the patient chart. The automated tool was axe-core 4.13.0.

The scan reported ten distinct problems under five success criteria. Examples included a help button with no accessible name, broken list structure in the user menu, date placeholders with a contrast ratio of 1.72:1 against a 4.5:1 minimum, and a link target only 16 pixels tall, below the 24-pixel minimum. The scan also left 165 checks marked incomplete, meaning axe could not decide on its own. The author calls that a to-do list, not a pass.

A manual pass with keyboard, 200% zoom, a 320-pixel viewport, Orca and TalkBack found seven more failures. Four concern focus: focus is lost when a table changes page, the calendar day has no visible focus style, focus is hidden behind a fixed bar at 200% zoom, and a clickable card header is a dead tab stop. The others were a registration form that is cut off at 320 pixels wide, a search result count that is not announced to screen readers, and a button nested inside a link, which axe did not flag in the author's test.

The author also describes using axe in continuous integration (automated checks on every code change) as a regression guard. A hand-written baseline file records accepted violations with a reason and date. New violations fail the build, and stale entries are flagged. The baseline is keyed by rule per page, so a second element breaking the same rule on one page would not be caught.

Why it matters

A clean automated report can give false comfort. According to the article, the scanner missed problems that depend on interaction, screen size, spoken announcements, or have no rule at all. Keyboard users met the focus failures first, and one failure made registering a patient impossible at narrow widths.

Automated scans are still worth running because they are fast and stop fixed problems returning. The practical lesson is to treat them as one layer, and to treat "incomplete" results as work for a person to review.

Source: DEV Community. This brief summarises their reporting; read the original for the full detail.