The say-do gap
RUCF measures what your product costs users, what they believe it costs them, and the distance between the two.
Behaviour is poor and nobody is complaining. Your users have stopped noticing what this costs them, which means it is invisible to every feedback channel you own.
Composite · illustrative
Score, then confidence. Drag the dot: the reading changes, the arithmetic stays published.
The problem
“How good is this product” produces a number nobody has to act on. Rate a checkout 3.4 out of 5 and you have described a feeling with a decimal point attached.
There are five arrangements of the same four dimensions in circulation. Usability, engagement, accessibility, satisfaction. They differ in wording and not much else, and none of them prescribes a different action depending on the answer.
RUCF asks three answerable questions instead. What did the user come to do. What did it cost them. What do they believe happened. Compare those and you get gaps, and the direction of a gap decides the treatment.
Behaviour and reported experience rarely agree. Most research practice treats that disagreement as noise to be resolved. RUCF treats it as the finding, because the direction of the disagreement decides what you should do.
Hatch · behaviour poor, perception good
Users have absorbed the cost, or they blame themselves for it. No survey will surface this and no support ticket will name it. It is also the cheapest quadrant to fix, because nobody is attached to a design they never noticed.
Treatment. Remove the step outright; explanations and training only preserve it.
Dotted · behaviour good, perception poor
Tasks complete at reasonable cost and people still call it confusing or slow or untrustworthy. This is the most commonly misdiagnosed state in the discipline, and rebuilding the interaction is the standard wrong answer.
Treatment. Change what the product says, not what it does.
Cross-hatch · both poor
Everyone already knows. Its information value is close to zero and its political value is high, which is why most research budgets end up here.
Treatment. Fix it, and recognise you are paying for confirmation rather than discovery.
Solid · both good
Behaviour and perception agree, and they agree in the direction you want. A quadrant to defend rather than invest in.
Treatment. Record it as a regression baseline and score every release against it.
State 01 / 04
The say-do matrix
Composite score
2.4@0.71
identical in all four states
Correct treatment
Remove the step outright
One composite score, four positions, four incompatible responses. This is the whole argument for measuring the direction of the disagreement rather than averaging it away.
Confidence
Measured. You measured this, recently, from behaviour. Act on it.
Guessed. Same score, different life. You have a survey from March and some strong opinions. Go and look.
Observed behaviour enters at 1.00. Reported experience at 0.70. Anything inferred, including AI classification of tickets or session replays, at 0.50. Expert judgment, including a senior designer’s heuristic review, at 0.30. That last figure is the correct weight. Everyone in the field knows it, and no other framework says so numerically.
Confidence also decays, and shipping is what triggers it rather than the calendar. Each release touching a measured surface multiplies its confidence by 0.8, because evidence describes a product that no longer exists. The loop closes itself: shipping a fix lowers confidence, which pushes that surface back up the measurement queue, so re-measurement follows without anyone needing to remember it.
Distributions
An average experience is a fiction. Most products work well for the majority and badly for a minority, and that minority is where churn, complaints, support cost and legal exposure all live.
One release, three readings
4.2
P50: the median session
1.8
P10: the worst tenth
2.4
Tail gap: P50 minus P10
RUCF reports the median and the tenth percentile. If they 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. Performance engineering stopped reporting mean latency around fifteen years ago for this reason. UX measurement has not caught up.
One surface, carried through the whole model. Each layer takes the output of the last and adds the thing the one before it could not tell you. Keep scrolling and watch the same reading change shape six times.
Layer 01
Intent, friction and perception, captured through separate channels and never allowed to inform each other while they are being collected.
Layer 02
The three readings become the corners of a triangle, and each edge is a gap. Effectiveness, expectation, and the awareness gap along the base that nothing else reports.
Layer 03
The awareness gap collapses the three readings into a single position on the matrix, and the position prescribes the treatment. This surface landed in silent friction.
Layer 04
The position carries the weight of the evidence behind it, and that weight falls by a fifth on every release that touches the surface. Three ships later it is worth half.
Layer 05
Recoverable cost times confidence, divided by effort. The smaller finding you observed outranks the larger one you assumed, which is the whole reason confidence is in the formula.
Layer 06
Shipping the treatment expires the evidence behind it, which raises the surface back up the queue on its own. The measurement schedules itself.
Layer 01 / 06
Three measures
The surface
Checkout
the same one throughout
Output of this layer
three readings, three evidence classes
The assembled model
Checkout costs 2.4, users call it fine, remove the step, re-measure in Q3.
The output is not a rating. It is a sentence with a treatment and an expiry date attached, which is the difference between a framework and a taxonomy.
Limits
RUCF structures user research rather than replacing it. It will not tell you what to build. It will not catch a broken business model or a product nobody wants. It is a poor fit below a few hundred active users, where the sample is too thin for the commitment signals to mean anything. It measures the cost of an experience and the accuracy of your beliefs about it. That is the whole scope.
Start with the free scorecard. If the result raises questions, book a session and we will run the full method against real behavioural data.