Framework · layer 3
The say–do matrix
Users tell you one thing. Their behaviour tells you another. Standard practice is to triangulate the two into a single truth and treat the discrepancy as measurement error.
That discrepancy is the most informative thing you have. Two products with identical usability scores need opposite interventions depending on which way their users’ perception is wrong.
The four states
Plot each surface on two axes
How much friction the behaviour shows, and how much friction users report. Each quadrant links to its own page with detection methods, treatment playbook and worked examples.
- Silent friction · high cost, no complaints
- Phantom friction · low cost, loud complaints
- Loud friction · high cost, loud complaints
- Flow · low cost, no complaints
- Watch · the dead band between them, where the reading is not yet reliable
Why direction beats level
Take two checkouts, both scoring 2.6
The first sits in loud friction. Users complain constantly, the team knows, three tickets are open. The information value of measuring it was zero. You are paying to confirm something the support inbox told you for free.
The second sits in silent friction. Nobody complains. It has never appeared in a survey, a ticket or an interview. Users have normalised the cost, or they assume the difficulty is theirs. This one is invisible to every feedback channel the company owns, and it is cheaper to fix, because nobody is attached to a design they never consciously noticed.
Same score. One of them is a discovery and one is a receipt.
Why perception-led research cannot find silent friction
Surveys, interviews and feedback widgets all sample perception. If perception is the thing that is wrong, the instrument reports that everything is fine.
This is not a failure of the practitioners. It is a structural property of asking people questions. The only way to find silent friction is to measure behaviour separately and then compare, which is what RUCF exists to make routine.
The prior art, honestly
The gap between stated and revealed preference has been documented in economics since Samuelson in 1938 and in social psychology since LaPiere in 1934. It is not new.
What is new here is treating it as the primary axis of a UX measurement model, attaching a distinct treatment to each direction of divergence, and tying the whole thing to an evidence confidence model that expires when you ship.