Stop reporting average UX. Performance engineering stopped fifteen years ago.
In three points
- An average experience is a statistical object nobody actually has. Report the median and the tenth percentile.
- The tail — the worst tenth of sessions — is where churn, support cost and legal exposure live.
- A wide gap between P50 and P10 is a design accusation: you built for people who resemble the team.
Around 2010, backend engineers stopped putting mean latency on dashboards. The mean was hiding the story: a service could average 200ms while one request in ten took four seconds, and those four-second requests were the ones users remembered, tweeted about and churned over. The discipline moved to percentiles — p50, p95, p99 — because the shape of the distribution was the finding, not its centre.
UX measurement has not caught up. We still average satisfaction, average task time, average effort scores, and publish the result as “the user experience” — a statistical object that no individual user is having.
The tail is a person
Most products work well for the majority and badly for a minority. That is not a failure of the products; it is the default shape of the distribution. The failure is reporting a number that erases the minority — because the minority is where everything expensive lives. Churn concentrates there. Support contacts concentrate there. Accessibility complaints, one-star reviews and, in the EU, legal exposure all concentrate there.
RUCF reports two numbers instead: the median session (P50) and the worst tenth (P10), with the difference between them named the tail gap. A 4.2 median with a 1.8 tenth percentile is not a good product with an edge case. It is two products sharing one interface — one designed, one endured.
What a wide tail gap accuses you of
If the two numbers sit far apart, you have not built a product with an accessibility issue. You have built a product for people who resemble the team that made it: same devices, same bandwidth, same fluency, same working conditions. The tail is everyone else, experiencing everything the demo never shows.
This is also why RUCF refuses to score accessibility as a dimension. Accessibility experience is not a column in a spreadsheet — it is the tail of every column. Conformance is a gate you pass or fail; the lived experience beyond conformance lives in the P10 of every signal you measure.
One honest caveat
A percentile needs a distribution, and a distribution needs segmented behavioural data. A single self-assessment cannot produce a real P10, which is why the scorecard returns a tail-risk flag rather than a fabricated percentile. A framework that invents the precision it argues for would be its own counterexample.
Signals this affects
The whole Access surface — contrast and legibility, input and assistive support, content clarity — plus the tail of every other signal.
Related: How to price friction in your own currency · What a scorecard cannot tell you