Reference documentation
The scoring model
Everything needed to run RUCF without us. The same anchors apply whether you run it by hand or through the scorecard.
Four surfaces, twelve signals
| Surface | What it captures | Signals |
|---|---|---|
| Task | Can people do the thing | Task completion, error recovery, navigation clarity |
| Commitment | Do they keep doing it | Activation, depth of use, return intent |
| Access | Can everyone do it | Contrast and legibility, input and assistive support, content clarity |
| Belief | Do they trust it | Visual hierarchy, brand coherence, trust signals |
Twelve rather than any other number because each surface maps to a distinct evidence source, and three signals is the smallest set that covers a surface without one signal carrying two ideas.
Signal definitions
Task completion. The share of users who begin the primary task and finish it in the same session. Evidence: funnel telemetry.
Error recovery. When something goes wrong, whether the user gets back on track without leaving or contacting support. Evidence: post-error session continuation, support contacts per error event.
Navigation clarity. Whether users know where they are and where to go next. Evidence: search use, back-and-forth loops, help visits before the primary task.
Activation. Whether new users reach first meaningful value. Evidence: cohort progression to a defined activation event.
Depth of use. Whether users go beyond the single entry feature. Evidence: feature adoption breadth per active user.
Return intent. Whether users come back. Evidence: month-over-month active retention.
Contrast and legibility. Whether text and interface elements are perceivable. Evidence: automated WCAG 2.2 AA scan of the main flows.
Input and assistive support. Whether flows work by keyboard, screen reader, touch alone, and on constrained devices. Evidence: manual keyboard traversal, assistive technology testing.
Content clarity. Whether interface language is understood without external explanation. Evidence: share of support contacts answering questions the interface should have answered.
Visual hierarchy. Whether the design directs attention to what matters. Evidence: first-interaction analysis, first-click testing.
Brand coherence. Whether the product uses one vocabulary and one visual language across interface, documentation, marketing and support. Evidence: terminology audit, whether users adopt your vocabulary back.
Trust signals. Whether users believe the product is safe, legitimate and honestly priced. Evidence: drop-off at payment, permission and personal data collection points.
Scoring criteria
Each signal scores 0 to 5 against published anchors, every anchor stating a countable range. The anchored answer options in the scorecard are the criteria — the tool and the manual method use identical anchors, so a hand-run score and a tool score are comparable.
Each signal is asked twice: once from behaviour (what the data shows) and once from perception (what users say), and every answer carries an evidence declaration — measured, estimated or guessed.
Composite
Behaviour drives the score. Perception drives the diagnosis. The composite is computed from behaviour only.
surface score = mean of its three behaviour signals
composite = Σ(surface score × weight) ÷ 100 × 5| Product type | Task | Commitment | Access | Belief |
|---|---|---|---|---|
| SaaS | 0.30 | 0.30 | 0.20 | 0.20 |
| Ecommerce | 0.35 | 0.15 | 0.20 | 0.30 |
| Content or media | 0.20 | 0.35 | 0.20 | 0.25 |
| Internal tool | 0.40 | 0.15 | 0.30 | 0.15 |
Adjust the weights if your product does not fit. Publish the weights you used alongside any score you report.
Score bands
| Band | Range | Reading |
|---|---|---|
| Failing | 0.0 – 1.9 | Users are completing tasks despite the product |
| Costly | 2.0 – 2.9 | Working, at a price you are probably not counting |
| Adequate | 3.0 – 3.7 | Nothing urgent, several things worth money |
| Strong | 3.8 – 4.4 | Optimisation territory, protect what works |
| Exceptional | 4.5 – 5.0 | Rare, and usually means a measurement error, so check confidence first |
Read the band with the confidence attached. A 4.6 at 0.30 confidence is a claim, not a result.
Reliability
Two people scoring the same product independently, both using observed evidence, should land within 0.4 on the composite. If your team is further apart than that, the disagreement is about the evidence rather than the criteria, and that disagreement is worth having before you continue.