Englyn, explained

Accessibility has been argued on moral grounds since before I had a career, and it has moved at roughly the speed you’d expect from an argument that asks people to be better than their quarter.

I’m not going to make that argument again. It’s correct and it hasn’t worked fast enough.

Here’s what changed. Something else now needs the exact same thing your screen reader users have needed since 1999, and it has a budget attached. Not because anyone got kinder. Because the machines can’t guess either.

The usual move here is to point at curb cuts. Built for wheelchairs, used by strollers and suitcases and delivery trolleys.

I’m going to skip it, for two reasons. It’s been said, including in the title of my last essay. And the more I sit with it, the more it reads like a quiet concession I don’t accept … that helping disabled people wasn’t reason enough on its own, and here’s a better one.

Here’s a measurement instead. A hundred and ten people, matched, half of them visually impaired. Three versions of the same site at three conformance levels, randomly assigned. Everyone did better on the more conformant build … and not disproportionately. Equally. No interaction effect at all.

The accessibility work wasn’t a tax with a side benefit. It was just better engineering, and somebody measured it.

In 2002 the W3C shipped a privacy standard called P3P. Sites declared a policy. Browsers checked a string was present. Nobody checked whether it was true.

Google’s policy string, in production, for years, read: “This is not a P3P policy!” The check passed.

Any standard you can satisfy by declaring something, somebody will.

And this isn’t a story about 2002. In July the W3C published WCAG-EM 2.0, its own evaluation methodology, and wrote into the normative text that using the methodology alone does not, in most situations, permit a WCAG conformance claim. The standards body saying out loud that its own method isn’t sufficient on its own.

It happened to me too, and that’s the honest version of this panel. Four automated layers reported clean on a page a screen reader was mispronouncing. Axe: zero violations. html-validate: pass. My own checker: pass. The DOM was correct.

And VoiceOver said “what this buys U-S,” because a CSS uppercase rule reaches WebKit’s accessibility tree. Six instances, four pages. Found by listening, and by nothing else.

Here’s the part I didn’t expect. The people who buy accessibility claims already know all of this. There’s no certifying authority for a VPAT … the body that owns the template says so plainly … so procurement reviewers interrogate the author instead of the claim. HHS asks what credentials the assessor holds. Massachusetts tells reviewers to research the firm.

Which means a report that marks a criterion CANNOT-CHECK, because it needs a real assistive-technology user and nobody ran one, is more credible under review than one claiming Supports across the board. The honest gap isn’t a philosophical flourish. It’s the thing that survives a reviewer.

Version 0.1.0 is a draft that publishes its limits in the same release as its claims. A standard about verifiable claims that hid its own would be self-refuting.

0 external reviews. No standards body, no independent audit. 1 assistive technology tested. NVDA, JAWS, TalkBack and Narrator are untested. 0 disabled reviewers have read it. The neurodivergent reading is first-hand; the screen-reader experience is not. Gold is undefined, deliberately, until the people it governs help write it.

If you build interfaces: run the checker against something you own and send me the report, especially if it fails. A FAILS from a stranger’s page is worth more to me than a hundred passes on my own.

If you use assistive technology: tell me where the specification is wrong.

Nothing about us without us is a demand, not a courtesy. The correct response isn’t to claim more. It’s to leave the door open and say which side of it you’re standing on.