Accessibility in assessment interfaces is a validity problem

The usual argument for accessible software is legal exposure. In assessment there is a stronger one, and it is the one that should carry the decision: an inaccessible interface changes what the test measures.

A candidate who cannot reach the next-question control is not demonstrating less knowledge. They are demonstrating that the control was unreachable. Any score derived from that session is measuring something nobody intended to measure.

Where assessment differs from the rest of the product

Most accessibility guidance assumes a user who can retry, ask for help, or come back later. A timed exam removes all three.

Time limits interact badly with assistive technology

A screen reader user moves through a question at a different rate than a sighted user, not because they read more slowly but because the interface serialises what a sighted user takes in at once. A time limit calibrated on sighted users is a different test for everyone else.

Error recovery has to be immediate

Outside an exam, an unclear validation message costs a few seconds. Inside one, it costs part of the answer. Messages need to say what to do, next to the field, without moving focus somewhere the candidate then has to find their way back from.

Things that are cheap and change outcomes

• A visible focus indicator that survives the design review.
• Controls that are real buttons and links, so they work the way the platform already taught the candidate they work.
• Status changes announced once, not on every keystroke.
• Contrast checked against the colour actually shipped, not the one in the palette.

None of these are difficult. They tend to be lost late, in the pass where someone removes an outline because it looked wrong on one screen.

Testing that finds real problems

Automated checks catch perhaps a third of what matters, and none of what matters most. The cheapest high-value test is to complete a full attempt using only the keyboard, then repeat it with the screen reader your candidates actually use.

# Useful, and not sufficient.
npx @axe-core/cli https://example.test/assessment/

If a keyboard-only attempt cannot be finished, no audit result changes that.

The part that is genuinely hard

Accommodation is not the same as accessibility. An accessible interface reduces how often an accommodation is needed; it does not replace the process for granting one, and treating it as though it does leaves candidates with no route when the interface still does not work for them.

Build both.