ADR 0025. The WCAG 2.2 AA audit is author-run

Status

Accepted. 2026-09-21. All sixty components in registry/bases/base/ were taken through a WCAG 2.2 AA audit, a static source pass and a rendered pass against the live /view page, and the findings were fixed in the component sources in the same change. REVIEW_FLOOR_NOTICE in lib/status.ts and rule SAFE001 in scripts/assert-ia.mts now read "Audited against WCAG 2.2 AA. Clinical review pending." rather than "no accessibility review and no clinical review." This record supersedes the review-floor paragraph of ADR 0023 and the "when the first accessibility review lands" condition of ADR 0024; the rest of both records stands.

Context

When ADR 0023 set the three release phases and ADR 0024 shortened the component page, no component had been checked for accessibility, so both records could say plainly that shipped meant the source installed and nothing more. That is no longer true. Every component has now been audited against WCAG 2.2 AA and the findings were fixed, so the flat sentence "no accessibility review" is false wherever it still appears. A fact that has changed cannot keep being stated as though it has not, and a design system for a health audience is the last place to leave a stale claim about what has and has not been checked. The question this record answers is exactly what the site may now say, and what it still may not.

Decision

The site says "audited against WCAG 2.2 AA" and says who ran the audit. It does not say "reviewed", "conformant", "accessible" or "WCAG 2.2 AA compliant", because those are claims about an assessment somebody else signs, and nobody has. The per-component box reads "Audited against WCAG 2.2 AA. Clinical review pending." Every machine surface and doctrine page that carried the flat "no accessibility review" now reads "no independent accessibility review", which stays true after an author-run audit. "Not for a production health surface" stays wherever it is a general statement, because clinical review is what gates that sentence and clinical review has not happened.

Consequences

The audit was run by the authors of the components. An author auditing their own work finds what they already knew to look for and is blind to what their own model of the component hides, so this is a floor raised, not a verdict delivered, and calling it "independent" would be the exact overclaim ADR 0023 was written to stop. Clinical review is still pending in full: no clinical safety officer, no hazard log, no clinical safety case and no data protection impact assessment exists for any component or screen specimen, and none is claimed. ADR 0023's premise is now partly historical. Its argument that alpha and beta claimed a difference nobody had measured still stands, but its observation that no component had had an accessibility review is now dated, and its "revisit when a component passes an accessibility review" condition has half fired: there is a measured audit, it is not independent, so a reviewed tier is still not earned and the three phases stay. ADR 0024's "revisit when the first accessibility review lands" condition has fired in its author-run form: the Accessibility section on each component page now reports audited results rather than a proposed bar, and the word budget held, so LEAN001 does not move.

Alternatives considered

A fourth release phase for "audited" was rejected. A phase word is a claim about code, an audit is a claim about a process, and wiring the two together is how "beta means accessibility reviewed" happened the first time; the audit rides in the box and in REVIEW_FLOOR_NOTICE, never in the phase. Rewriting ADR 0023 and 0024 in place was rejected as well. They are dated records of what was decided when, and editing their premises would erase the history this record exists to add to; each keeps its text and gains one pointer line to here. Claiming conformance rather than an audit was rejected because conformance is a statement an independent assessor makes, and substituting the authors' own audit for it is the substitution this whole record refuses.

Revisiting this

Revisit when an independent accessibility audit is commissioned, because that is the event that lets the box drop "author-run" and say a second party checked the work. Revisit separately when a clinical safety officer signs a safety case, because that is the event that lets a page drop "Clinical review pending" and speak about a production health surface. Neither has happened, and until one does the sentence stays exactly as it reads now.