The published site has had no human accessibility pass

[ChelseaKR]

The README states the boundary of the automated checks directly:

What still needs a person: none of this looks at the pages. Layout, reflow at small widths, focus visibility in practice, and reading order under a real screen reader are not settled by any check here.

The automated floor is genuine and honestly scoped — contrast is measured as arithmetic over both palettes in tests/test_site.py, page structure is checked in both toolchains so make verify keeps a floor when node is unavailable, and rules that jsdom cannot run (target-size, contrast) are reported as not-run rather than as passed.

But the site is a build artifact published from main by pages.yml, and nobody has looked at it with a screen reader.

Done when: a person has worked through the published pages — layout, small-width reflow, focus visibility, reading order — and the result is recorded.

Imported from the README ("What still needs a person") with AI assistance.

[ChelseaKR]

Triage against origin/main (8ae3fa2): still open, premise intact, not automatable.

The automated floor is real and I confirmed it is still in place: make pages runs html-validate plus axe-core over the built pages, tests/test_site.py computes contrast as arithmetic over both palettes, and rules jsdom cannot execute (target-size, contrast) are reported as not-run rather than as passed — which is the correct shape and the reason this issue is honest rather than alarmist.

What is missing is exactly what the README says: nobody has opened the published pages with a screen reader. Layout, reflow at small widths, focus visibility in practice, and reading order under a real AT are not decidable by any check in this repository, and an agent asserting them would be manufacturing the evidence the issue exists to say is absent.

Reported as owner-blocked: it needs a person and a recorded result. Not closed.

Triaged with AI assistance.

[ChelseaKR]

Labelled good first issue, help wanted and accessibility, with the scope this label is claiming.

The published site is five documentation pages — index, gates, california, action, evidence — not an application. No forms, no scripts to drive, one long table. Roughly 10 to 20 minutes per page, more for gates because a long table read linearly by voice is where reading order and header association usually break.

One page is a whole contribution. Walking one and stopping is expected, not a partial job, and that is why the label is defensible here where it would not be on a whole-application walkthrough.

The most likely real finding is on the gates page, and the second most likely is focus visibility, since tests/test_site.py measures contrast as arithmetic over both palettes but nothing here has ever looked at a rendered focus ring.

A session-report template scoped to one page is proposed in #62, along with docs/HELP-WANTED.md. This issue stays open until somebody has actually read the pages.

Prepared with AI assistance; reviewed before submission.