71.21 percent is 47 out of 66
Carrefour's accessibility statement said 71.21 percent. Under the RGAA method only one fraction gives that number, and it says less than it seems to.
Imagine two online shops, both audited with the French RGAA method, both with 100 applicable criteria.
The first passes 99 of them. The one it fails is keyboard access, on the button that says "Pay." The second passes 57 and fails 43, all of them small enough that nobody is stopped from buying anything.
The first shop's accessibility statement says 99 percent, the second says 57, and both say "partiellement conforme." Only one of them sells anything to a customer who doesn't use a mouse.
In June a court in Caen read a real number off a real statement, Carrefour's: 71.21 percent. The judge used it against Carrefour. I wanted to know what is inside a number like that, so I worked it backwards.
Where the number comes from
The statement is based on the RGAA, the French method for checking conformance with EN 301 549. Its conformance rate is the number of criteria validated, divided by the number of criteria applicable.
Two rules sit inside that. A criterion counts as validated only if it passes on every page of the sample, and on a single page one failing element is enough to fail it. A criterion counts as applicable if it applies on at least one page. The ones that apply nowhere leave the denominator.
The RGAA has 106 criteria, so the denominator is at most 106. I checked every fraction with a denominator up to 106. Exactly one rounds to 71.21 percent: 47 out of 66.
So 66 criteria applied to Carrefour's sample. 47 passed everywhere, 19 failed somewhere. The order from Caen lists 19 non-conformities, from text alternatives to keyboard access to visible focus. The arithmetic and the court agree.
What does 71.21 percent count?
Criteria. Not barriers, not pages, not people.
One of the 19 can be a single image with a poor text alternative on a page few people visit. Another can be a checkout button nobody can reach with the keyboard. Each is one nineteenth of the failure. For a customer, one of them is a typo and the other is the end of the purchase.
The denominator moves too, and that is where it gets a little absurd. Add a video, and the criteria for captions and audio description become applicable. Done well, the rate goes up. Done badly, it goes down, although nothing that was already there got worse. Delete a form, and the form criteria leave the fraction, and the rate may well improve. The RGAA says the rate is there to "mesurer les progrès du service en ligne," to measure progress. Deleting the contact form is not what anyone means by progress.
Strict inside, generous across
The method is strict inside a criterion and generous across them. Inside, it takes the minimum: one failing element on one page, and the criterion fails. Across, it takes the mean: 47 passes and 19 failures make 71.21 percent, as if the 19 could be paid off with the 47.
A purchase doesn't take the mean. It is a path, every step has to work, and one step at zero makes the whole path zero. I call that the one-zero rule, mostly because I keep needing it. The first shop from the beginning is the clean case: a mean of 99 percent, and a minimum of zero exactly where the money is.
Partial failures don't average either. If each of eight steps loses five percent of some group of users, independently, 0.95 to the power of 8 leaves 66 percent of them at the end. If the steps lose different people, it's 60 percent. The average step still says 95.
Same words for 50 and 99
The accessibility statement turns the rate into words. Full conformance if all criteria are met. Partial conformance if at least 50 percent are. Non-conformance below that, or if there is no valid audit. So 50 percent and 99 percent get the same words, which is how the two shops ended up with the same label.
The thresholds are easy to misread, even for the regulator. In a decision from 24 June 2026, Arcom found that a ministry service declared full conformance with a measured rate of 42.11 percent. Arcom wrote that below 50 percent the statement should say partial conformance. Earlier in the same decision it had quoted the RGAA correctly: below 50 percent is non-conformance.
What the standards say about scores
WCAG 2.2 has no percentage. A page conforms at level AA if it meets all level A and AA criteria. WCAG-EM 2.0, the W3C's evaluation method from July 2026, allows an aggregated score as an optional step and then warns about it: "aggregated scores can be misleading and do not provide sufficient context and information to understand the actual accessibility of a digital product."
The WCAG 3 draft from 10 September keeps conformance binary: "all core requirements MUST be satisfied." Percentages appear only as an idea for reporting progress between tiers. Requirements get severity tags instead, from Physical Harm to Friction, which is closer to the difference between the typo and the checkout button.
(Lighthouse is the other number people quote. Its accessibility score is a weighted average of automated audits, and Manuel Matuzović built an inaccessible site with a perfect score in 2019. Same problem, fewer criteria.)
What the judge did with it
The judge in Caen did none of this arithmetic. He wrote that the site "ne peut pas être seulement un peu accessible, il doit l'être totalement": it can't be a little accessible, it has to be fully accessible. For a binary standard, that is the correct reading of 71.21 percent. It is a list of 19 failures with a number in front of it.
So, to whoever signs their company's accessibility statement: put the fraction next to the percentage, and the list of what failed under it. The court in Caen read the list. Your customers won't read either. They run into the list.