Fictional company (Kestrel Supply Co.). SYNTHETIC data. ILLUSTRATIVE prices. Proof of concept.

Accessibility

This page targets WCAG 2.2 AA, with the limits listed below. It does not claim conformance. Two independent audits have been run on it (see "The independent audit, 2026-10-28" and "The second independent audit, 2026-10-29" below). No screen reader has been used on it, and nobody who uses assistive technology has reviewed it.

Applies to the six pages in site/: index.html (the main page), case-study.html, memo.html, failures.html, accessibility.html (this page, made from this file by a script) and sitemap.html (the site map, which lists every page and what it is for). The table of criteria was written for the main page. The other five are plain pages of text, lists and tables, and what is checked on them is listed under "The other pages". Kestrel Supply Co is fictional and the data is synthetic.

How to read the table

• Status: met, not met or not applicable. This is my reading of the page as built.
• How verified:
  • automated: a check that runs on the whole project fails if the thing breaks. Each check has a broken fixture that proves it can fail (tests/sitechecks.test.mjs).
  • manual: I checked it by hand, in the built-in browser pane of the Claude desktop app, or by reading the page text and code. A person did not do this, and no assistive technology was used.
  • not verified: nobody has tested it. The status is only my expectation from the markup.
• automated + manual means both were done.
• WCAG 2.2 removed criterion 4.1.1 (Parsing), so it is not listed. A to AA in 2.2 is 55 criteria: 31 at level A and 24 at level AA.

The independent audit, 2026-10-28

On 2026-10-28 the built pages were audited against WCAG 2.2 A and AA. The audit was done by Claude, an AI assistant, in a fresh chat that had not seen the build. It used headless Chromium and axe-core 4.10.2. It was not done by a person. It did not use a screen reader, another browser or a real device, so it says nothing about those.

Of the 55 Level A and AA criteria, the audit found 34 met, 5 not met, 16 not applicable and 0 it could not verify. The five not met were 2.4.3, 1.4.10, 1.4.11, 1.4.12 and 2.4.5. Each is recorded below as not met when it was found. Four were fixed on 2026-10-28 and the fifth, 2.4.5, on 2026-10-29. Each fix began with a check that failed on the page as it stood (the failures are in the technical log).

Found in the same audit, lower priority, and not recorded as the failure of a criterion. All four were fixed on 2026-10-28, each after a check that failed first:

• The three tables in the case study had no caption. They have one now, and a check requires a caption on every table on every page.
• Two regions on the main page had the same name. The new check for this found two pairs: the two lists of page sections were both called "On this page", and one table for Chat assistant A had a twin in the worked example. Each now has its own name, and a check fails if two landmarks or regions on a page share a role and a name.
• Long chart descriptions were read out in full as the name of the chart. A chart is now named by its short title and described by its description, and a check requires that.
• Nothing announced that the data had changed when the reader switched data sets. The "what has moved" message now says which data is showing, and a browser check reads it.

2.4.5 Multiple Ways was the auditor's finding, not mine.

The audit marked 3.2.3 Consistent Navigation as met, with a note that the forms of the navigation differ. This project's own list had marked it not applicable, because the notes described one page. One navigation line, the same on every page, was added on 2026-10-28.

Where the audit and this project's own list differ

The audit and this project's own list gave different results on these eight rows. The "before the audit" column is this project's list as it stood in the notes before the audit.

The audit's result for each of the 55 criteria was given to me as a list, and these eight rows are every difference between it and this project's list before the audit. This project's list still has 2.2.2 and 3.1.2 as not applicable, because nothing starts moving by itself and all the text is English. The audit marked both met. I have not changed either row, and the table shows the difference and does not settle it.

The second independent audit, 2026-10-29

On 2026-10-29 a second audit was done by Claude, an AI assistant, in a fresh chat. It used headless Chromium, axe-core 4.14.0 and Lighthouse 13.6.0. It was not done by a person.

Its scope was index.html only. It did not look at the case study, the decision memo, the failure log, this page or the site map.

Of the 55 Level A and AA criteria, it found 34 met, 0 not met, 20 not applicable and 1 it could not verify. The one it could not verify is 4.1.3 Status Messages, which needs a screen reader.

This is a result for the main page only. It does not say that the whole site has no criterion not met. For the other pages, the first audit and the checks listed under "The other pages" are what there is.

The second audit's totals differ from this project's list by more than 4.1.3. The audit's row-by-row list was not available when this was written, so the other rows are not named here. This project's own list, written for the main page, is in the tally below: 37 met, 0 not met and 18 not applicable. It is not the same list as the second audit's 34, 0, 20 and 1. Nothing in the criteria table was changed because of the second audit, since it found no criterion not met.

Criteria

Tally

55 criteria: 31 at level A and 24 at level AA.

There are two lists, and they are not the same list.

The audit's list, as the audit gave it, before any fix: 34 met, 5 not met, 16 not applicable and 0 could not verify.

This project's own list before the audit: 35 met, 0 not met and 20 not applicable.

This project's own list now, counted from the criteria table above:

Of the 37 criteria marked met:

So 22 have an automated check and 24 have a manual check. The 18 not-applicable rows were each confirmed by a manual look for the content in question. A check reads the criteria table and fails if these counts do not match it.

The other pages

The case study, the decision memo, the failure log, this page and the site map are plain pages of text, lists and tables. They have no script and no outside file. These are checked by the same automatic checks:

• Each has the notice that the company is fictional and the data synthetic, no outside reference, no em dash and no hype word.
• No cut-off, shortened or overlapping text at 1280, 640 and 320 wide, and at 320 wide with the four text spacing values and a wide font. No sideways page scroll at 375 and 320 wide.
• Every table has a caption, sits in a labelled region that scrolls with the keyboard, and has a scope on every heading.
• The words on the case study, the failure log and this page are the words of the file each is made from, and the page is what the script makes. A hand edit fails.
• Each has the same navigation line at the top, and links to every other page. The site map lists every .html file in site/, with a line on what each is for, and fails if one is missing.
• The memo prints on exactly one page.

Not checked by a program: the contrast of text on these pages, the order of their headings, and the names of their links. They use dark text on white. They have not been used with a screen reader.

The pixel animation

The café in pixels is a second view of the counter picked above. Rows marked "verified by Chris, Edge on Windows, 2026-10-28" are what he watched in a real Edge window on Windows. His first look on that day showed only the still picture, because Windows animation effects were off and the page correctly followed that setting. He turned animation effects on and then watched the animation. Before that, an earlier version of this section marked rows as verified when he could not yet see it play. That was wrong and is corrected here (see the technical log). Every row not marked as verified is CSS-verified only: to be verified by Chris by hand. An automated check reads the stylesheet and the page, and a browser test sets the animation clock by hand in a headless browser. They are not counted in the tally above, which covers the 55 criteria for the whole page and has not changed.

Limits

What has not been tested, and what that means:

• Screen readers. None used, by me or in either audit: not NVDA, JAWS, VoiceOver, TalkBack or Narrator. Names, roles and the live region are checked in markup only.
• The independent audits. Both were done by an AI assistant in a fresh chat, with headless Chromium: the first with axe-core 4.10.2, the second with axe-core 4.14.0 and Lighthouse 13.6.0. They are second looks and not expert reviews. Neither used assistive technology, another browser or a real device. The second looked at the main page only.
• Chart text at narrow widths. At 320 px wide the labels in the charts shrink to about 9.8 px, and they do not grow when only the text size is raised. Every chart has a table version, which does grow. That is the second audit's figure. My own measure of the built page, in a window 320 px wide, is 9.3 px in the seats chart. Raising only the text size makes the page reflow to one column, and the charts then get smaller, not larger.
• The data set choice. The data set radio buttons redraw the whole page when they are changed. The page says which data is now showing in the status message, and focus stays on the choice (a browser check reads both), but the cards, the café and the seat list all change at once.
• Screen readers, other browsers and real devices. Screen readers, other browsers and real devices are untested by either audit. The only real browser window used on the page is Edge on Windows, by Chris, for the rows marked as verified by him.
• Other browsers and devices. Manual checks used one embedded Chromium-based browser pane. Not tested in Firefox, Safari, Edge, or on a phone or tablet. The 320 px and 200% tests resize that pane. They are not a real device at 400% zoom.
• Forced colours. Verified by Chris, Edge, 2026-10-27. He scrolled the whole page with forced-colors: active emulated in the Edge DevTools Rendering panel, and nothing broke. A check also confirms the stylesheet has a forced-colors block.
• Prefers contrast and OS text scale. Checked by Chris in Edge DevTools emulation, 2026-10-27. He scrolled the whole page with prefers-contrast: more and with OS text scale at 200%, and nothing broke.
• Font-size-only 200% overflow (known limit). If only the page's base font size is set to 200%, with no change to the window or the breakpoints, some tables in the tool cards still scroll sideways at 1280 px wide. Not reproduced by Chris under OS text scale emulation. I judge a real text size setting also changes the layout breakpoints, but I have not checked that in Firefox.
• Reduced motion. Verified by Chris, Edge on Windows, 2026-10-28. With Windows animation effects off, the page showed the still picture, no Play button and the explanatory note. With them on, the Play button appeared and the picture was paused until he pressed it. A check also confirms the stylesheet has the rules, and a browser test forces the setting on and off. Both browsers and the in-app pane on the machine I used ask for reduced motion by default, so I tested the animation itself with those rules taken out of a test copy. Edge on Windows is the only real window it has been seen in.
• Voice control, switch access and magnifiers. Not tested.
• Wording quality. Text alternatives, link names and headings are checked for presence and structure. I judged the wording. Nobody who uses assistive technology has reviewed it.
• Plain English. The reading-grade check measures sentence length and syllables, not whether a term has been explained at first use. A jargon check covers a short list of words. See the technical log.
• No JavaScript. All reading content and the section links work without it. The sliders, data switch and seat filters need it, and the page says so.
• Not covered by the checks. Any problem the checks do not look for, which is the usual limit of automated testing.

No criterion in the table above is marked "not met" now. A criterion marked "met" means I found no failure with the checks, the tests and the audits above. That is not the same as none failing.

How the checks are run

The checks run on the whole project, and any one of them failing stops a change. The checks are WCAG 2.2 AA static checks, WCAG 1.4.11 non-text contrast (a list of pairs, and every chart mark), colour contrast, plain English, every id is unique and every link ... exists, the layout and page checks, the interaction probe, and the example and sidebar checks. The manual items need a browser: open the pages in one.