A Lighthouse accessibility alternative

Lighthouse's accessibility audits come from axe-core, the engine this scanner also runs, but Lighthouse uses a subset of axe's rules. The score is a weighted average of those audits, and each audit is simply passed or failed. One unlabelled field and fifty unlabelled fields fail the same audit and cost the same points.

That makes the score useful as a trend and poor as a to-do list. A page can sit in the 90s with a checkout form nobody can use by keyboard, because Lighthouse cannot test that and says so: it lists "additional items to manually check" under the score.

Every failing element, grouped by severity, with the WCAG success criterion and Level A or AA next to it. Findings that need a person are kept in their own list rather than folded into a score.

• Resolved contrast. Text on a CSS gradient comes back from axe, and so from Lighthouse, as "needs review". This measures the worst-case colour stop and gives a real pass or fail, as explained in resolving contrast over CSS gradients.
• An export you can track. Download the findings as CSV, JSON or Markdown and work through them as tickets.
• A build gate. The GitHub Action fails a pull request that adds a violation, which a score threshold cannot do per issue.

Both tools test the machine-checkable part of WCAG, which is roughly a third of the success criteria. A clean automated result means the checkable part passed. Keyboard access, focus order and meaningful alt text still need a manual pass, whichever tool you use.