Accessibility is a gate, not a score
In three points
- WCAG conformance is a pass or fail with legal force in the EU, so RUCF removes it from the composite entirely.
- Fail the gate and the framework returns a remediation list instead of an index. There is no number to negotiate with.
- Accessibility experience is a different thing from conformance, and it lives in the tenth percentile of every dimension.
Almost every UX scoring model treats accessibility as a dimension. It sits alongside usability, visual design and content, receives a number out of a hundred, and is averaged in with the rest. The result is a product that fails a statutory obligation and reports 78 out of 100, because the three dimensions it is good at carried the one it is not. That is not a rounding problem but a category error, and RUCF removes conformance from the score.
Fig. 01
A gate, and a tail. Two different mechanisms
What averaging a legal requirement actually does
WCAG 2.2 defines testable success criteria at three levels, and Directive (EU) 2019/882 gives conformance legal force for in-scope products and services in the European Union. A criterion is met or it is not met. There is no partial credit in the standard and no partial credit in the directive, so any weight below 1.0 applied to conformance is an invention of the framework rather than a property of the thing being measured.
Give it a weight of a quarter alongside three other dimensions and here is what you have built: a mechanism by which an attractive, well-written, fast product can break the law and still print a respectable figure on a slide. The figure is persuasive. It gets circulated, it anchors the roadmap conversation, and the twenty-two point deficit hiding inside it becomes a line item competing with a marketing-site redesign. A composite in that position is not neutral. It launders an obligation into a preference.
So the gate runs first
In RUCF, conformance is evaluated before the composite exists. Any unmet level A or AA criterion on a measured surface fails the gate, and a product that fails the gate does not receive an index at all. It receives a remediation list: the criteria, the surfaces, and the order to fix them in. There is no number to argue with, because a number invites the argument that this year the number is good enough. A team that wants its composite has to clear the gate first, which is a stronger incentive than any weighting scheme.
One caveat travels with the gate: automated scanning does not establish conformance. Depending on the tooling and the criteria in scope, a scan detects somewhere around a quarter to a third of what the standard requires, and it is silent on almost everything involving meaning, order and alternative text quality. Keyboard traversal, focus order, screen reader output and reflow at 320 pixels are checked by a person, at a desk.
The second thing, which is not the first
Conformance and accessibility experience are different quantities that share a word. A product can satisfy every success criterion and still be miserable to use. The label is present and correctly associated, and it reads as forty-one characters of legal boilerplate. The focus order is logical and requires nineteen tab presses to reach the primary action. Both pass the standard. Neither is acceptable.
That experience does not belong in a dimension of its own, because a dimension can be averaged away by the dimensions beside it, which is how the problem was created. It belongs in the tail. RUCF reports every measure at the median and the tenth percentile, and accessibility experience is what the tenth percentile is made of: the screen reader session, the session on a three-year-old Android on a congested network, the session by someone who needs the text at two hundred per cent and finds the layout collapses at one hundred and eighty.
Conformance is closed by remediation: specific defects, specific criteria, done or not done. Tail experience is closed by design, which is never finished. One number forces a single treatment onto two problems, and the cheaper treatment always wins.
What this does not solve
The gate can be gamed by scoping. Declare three surfaces in scope, pass, and leave the settings area unexamined. RUCF has no defence beyond naming the scope in the output, which at least makes the omission legible. The framework does not perform the audit and does not produce a legal opinion. It states where conformance sits in the arithmetic, which is outside it.
Signals this affects
The Access surface: contrast and legibility, input and assistive support, content clarity, plus the tenth percentile of every other signal in the model. The gate itself sits outside the score, in the framework.
Related: Stop reporting average UX · What a scorecard cannot tell you