---
name: accessibility-regression-test
description:
  Converts a verified accessibility repair into the narrowest durable
  user-behavior test. Use after an accessibility fix or when asked for an axe,
  component, Playwright, keyboard-journey, or manual accessibility regression
  record.
license: MIT
metadata:
  version: "1.0"
---

# Accessibility regression test

Protect the user-facing behavior that failed, after the repair has been verified
in the original state.

## Workflow

1. Read the before evidence and verified after evidence. State the barrier in
   terms of user behavior.
2. Choose the narrowest reliable layer with
   [references/test-selection.md](references/test-selection.md).
3. Make the test fail against the known broken behavior when practical, then
   pass with the repair.
4. Assert outcomes such as focus destination, exposed error, accessible name, or
   absence of the original fingerprint. Do not assert only that an attribute
   string exists.
5. Control route, state, viewport, reduced motion, and data so the test is
   repeatable.
6. Keep manual-only outcomes as an explicit review record rather than a
   misleading automated test.

## Boundaries

Do not snapshot raw axe output, browser versions, run IDs, or timestamps. Do not
weaken a rule, baseline away a new issue, hide content, or remove functionality
to make a test pass. A regression test protects its scoped behavior; it does not
certify the component or page.