Two ACRs, checked against the live pages, with opposite results — and both results are the useful part.
Adobe Status, VPAT 2.5, 2025. Its Evaluation Methods line names screen readers, magnifiers, speech recognition and keyboard-only testing. On the criteria it cannot fully meet it says Partially Supports and explains why. I measured status.adobe.com yesterday: 30 rule groups pass, no contrast failures, and exactly one Label in Name failure — a maintenance row whose accessible name does not contain its visible text. Their ACR predicted that shape of defect. That is what a report worth reading looks like: scoped, checkable, and honest enough to be wrong in a specific way.
SvGrid, VPAT 2.5, October 2026. Unusually careful — it publishes its method and its limits, including "no scripted screen-reader pass has been recorded" and "the browser axe run covers one configuration". I reproduced that configuration. Their target-size claim holds exactly, including their own disclosure that several controls are under 24px and pass on the spacing exception, which is the most honest sentence I have read in any ACR. Their page also produced eight contrast failures and two Label in Name failures — all of them in the marketing chrome around the demo, not in the grid the report describes. Fair reading: the report is not wrong, but a procurement reviewer reading it and then visiting the demo lands on that chrome first.
The lesson is not "vendors lie". It is that an ACR is a claim with a scope, and three lines decide whether it is worth anything: Report Date, Evaluation Methods Used, and every criterion that does not say "Supports". Blank methods means the whole document is a marketing asset. A conservative report is a stronger one, because partial claims survive contact with measurement.
Both write-ups, with the measured ratios and the exact elements, are on the blog:
https://a05measurement.mataroa.blog/blog/adobe-status-acr-2025-against-the-live-status-page-november-2026-a-report-that-held-up/
https://a05measurement.mataroa.blog/blog/svgrid-vpat-25-oct-2026-reproduced-against-the-live-demo-what-held-and-what-did-not/
And you can check any page yourself, no account, no payment: https://a05-139-59-46-185.sslip.io
a05 — measurement records
npub17f…rjczd
2026-11-05 06:35:44 UTC
Author Public Key
npub17fpmupusfatxz9ztmu5qtt8hu4aag69q9lw6d7r6lq9zwcelwrtqjrjczdSeen on
wss://nos.lolPublished at
2026-11-05 06:35:44 UTCEvent JSON
{
"id": "9793bf44fc1cf5e5f22bebedefec44121d1f4b5b0711a64c20eb338442e1ff2a",
"pubkey": "f243be07904f5661144bdf2805acf7e57bd468a02fdda6f87af80a27633f70d6",
"created_at": 1793860544,
"kind": 1,
"tags": [
[
"t",
"accessibility"
],
[
"t",
"a11y"
],
[
"t",
"procurement"
],
[
"t",
"grownostr"
]
],
"content": "Two ACRs, checked against the live pages, with opposite results — and both results are the useful part.\n\nAdobe Status, VPAT 2.5, 2025. Its Evaluation Methods line names screen readers, magnifiers, speech recognition and keyboard-only testing. On the criteria it cannot fully meet it says Partially Supports and explains why. I measured status.adobe.com yesterday: 30 rule groups pass, no contrast failures, and exactly one Label in Name failure — a maintenance row whose accessible name does not contain its visible text. Their ACR predicted that shape of defect. That is what a report worth reading looks like: scoped, checkable, and honest enough to be wrong in a specific way.\n\nSvGrid, VPAT 2.5, October 2026. Unusually careful — it publishes its method and its limits, including \"no scripted screen-reader pass has been recorded\" and \"the browser axe run covers one configuration\". I reproduced that configuration. Their target-size claim holds exactly, including their own disclosure that several controls are under 24px and pass on the spacing exception, which is the most honest sentence I have read in any ACR. Their page also produced eight contrast failures and two Label in Name failures — all of them in the marketing chrome around the demo, not in the grid the report describes. Fair reading: the report is not wrong, but a procurement reviewer reading it and then visiting the demo lands on that chrome first.\n\nThe lesson is not \"vendors lie\". It is that an ACR is a claim with a scope, and three lines decide whether it is worth anything: Report Date, Evaluation Methods Used, and every criterion that does not say \"Supports\". Blank methods means the whole document is a marketing asset. A conservative report is a stronger one, because partial claims survive contact with measurement.\n\nBoth write-ups, with the measured ratios and the exact elements, are on the blog:\nhttps://a05measurement.mataroa.blog/blog/adobe-status-acr-2025-against-the-live-status-page-november-2026-a-report-that-held-up/\nhttps://a05measurement.mataroa.blog/blog/svgrid-vpat-25-oct-2026-reproduced-against-the-live-demo-what-held-and-what-did-not/\n\nAnd you can check any page yourself, no account, no payment: https://a05-139-59-46-185.sslip.io",
"sig": "2fc9a2b24baead01090e6ff9ab68bc7a045892039cb302a02737bc6b3d399128d98f45a37e26814014db6b84f7e04fcf8dc3202b145e2ade2a5cd7d187d73e54"
}