The record

What was asked

Twenty-eight dimensions, in six families: what the work is, what it says, what it claims, what a reader can do with it, whether it behaves, and the floor.

Bounds, declared before anything was read

• 01To pass, the average had to reach 8.0.
• 02To pass, no single dimension could fall below 6.0.
• 03The budget was five dollars. A little over forty cents was spent.

How it was read

Eleven separate reads. Six were declared before the bar was frozen, so nobody could pick convenient ones afterwards. Five more were added when the first set left gaps — and the fact that five were missing is itself part of the record.

The declared six

• 01Every link, followed. Failed — four anchors in the navigation pointed at sections that do not exist. Three are now fixed.
• 02Accessibility, WCAG 2.2 AA. Failed — contrast, on all three pages.
• 03Page weight, request count, script count. Neither — the caps were never given numbers, so the guard could not fire. A threshold nobody declared is not a threshold.
• 04Sitemap and robots. Passed.
• 05Every word against the brand's never-use list. Passed.
• 06Every published claim traced to a retained source. Passed.

Five judges that were not the author

Lines that need a person to judge were given to five models of five different families. None of them wrote the copy. None of them was told which page was a deliberately bad control.

All five caught the control, so all five kept their vote. On thirty cells — ten lines across three pages — nine passed, two failed, seven could not be observed, and twelve came back ambiguous.

Ambiguous means no four of the five agreed. The protocol records that as a finding about the bar rather than about the pages: a line that independent readers cannot agree on is not yet specific enough to be judged by readers at all.

Two scored runs, because one is a sample

The artifact was scored twice, on identical bytes. The two runs disagreed by as much as 2.5 on a single dimension, and by 0.725 on average.

That number matters more than any score here. It is the floor below which movement is not movement, and it is why a claim of improvement smaller than it would not be recorded as an improvement.

What it found

• 01Accessibility, 2.0. Contrast failed on all three pages. This one is not fixed yet.
• 02Instrumentation, 1.0. Nothing on the site declares what happens when a reader acts. The hosting platform records visits; the pages say nothing about it, which is its own problem.
• 03Cost of acting, 2.0. No action on this site states what it costs you. Every action here is free — and none of them says so.
• 04One claim per paragraph, 3.5. The custody page carries several claims per paragraph. Five readers disagreed about four of those lines, which turned out to be a disagreement about the word "claim" rather than about the page.
• 05Links to a section landing on that section, 2.0. The same four dead anchors the link read found. Two independent reads, one defect.

And the strongest lines

Purpose clear in five seconds scored 9.25 and passed on all three pages, unanimously. Sitemap and URL stability scored 9.5 and 9.75.

Why the verdict is "needs owner" and not "rejected"

Three dimensions had no observation at all. Nothing measured whether a reader can complete a task, whether one action per page is reachable in one click, or how the pages load on a real device.

A dimension nothing observed cannot be scored as a pass or a failure. It routes to a person. That is the rule, and it holds regardless of what the other twenty-five dimensions said.

The consequence is uncomfortable and worth stating: until those things are measured, this site cannot pass its own bar. It can only keep improving and keep returning to a person.

What the instrument got wrong

The apparatus that produced this record is not the same apparatus that started. Eight defects were found in it while it ran, and every one was the same kind of mistake: a single observation, or a hand-written pattern, treated as complete.

• 01A dimension count asserted rather than computed — the wrong number was reported before it was checked.
• 02A check that counted sentences and called them claims.
• 03A link reader that dropped every relative link, and reported that a page with a navigation had none.
• 04A word-list check that found no list, then reported a pass.
• 05A panel run that wrote its results only at the end, was killed by a timeout, and lost every reader's answers — a full run bought twice.
• 06A reader that dropped four of twenty-eight dimensions, one of them invisible to its own pattern — and passed its own completeness check.
• 07A probe that never passed the address to the thing it was probing, and reported five failures it had not measured.
• 08A verification that fetched a page once and called it stale, inside the window before a deployment propagates.

None of these was a defect in the pages. Every one was a defect in the instrument, and every one was caught by running the check and disbelieving the result — never by reading the code.

The honest summary is that a record is only as good as the instrument that produced it, and the instrument required eight repairs on its first outing. That is in the record because a record that omitted it would be the thing this site is against.

What is still wrong with this record

• 01The same agent wrote the pages, wrote the bar's instrument, and ran the loop. A maker grading its own deliverable is the failure the whole apparatus exists to prevent. The mitigation is that nothing scored the pages except machines of other families — and that is a mitigation, not a fix.
• 02One anchor is still dead. It points at a section that does not exist, and no amount of good work fixes a link to nothing.
• 03Twelve lines of the bar are not specific enough for independent readers to agree on. That is a defect in the bar, and it is recorded here rather than quietly averaged away.