Each of the N test users is placed on a linear onboarding flow of screens S₀ → S₁ → … → Sk (goal). At every screen, a user's next move is drawn from a simple 3-outcome model with per-screen error probability p (the slider):
P(advance) = 1 − p
P(backtrack) = 0.6 p (confused, retries the previous screen)
P(abandon) = 0.4 p (gives up — a real "usability failure")
This is the same task-based methodology described in the source report: users attempt a scripted flow and the run is scored with the standard formulas —
Completion Rate = successes / total users × 100%
Mean Time-on-Task = mean(steps taken by completed users) × 6s/step
The SUS (System Usability Scale) readout is a simplified illustrative regression that tracks the well-documented correlation between task completion rate and SUS scores (Sauro & Lewis), not a substitute for a real 10-item SUS questionnaire:
SUS_est = clamp( 68 + 0.5×(CompletionRate − 70) − 10×meanBacktracksPerUser, 0, 100 )
- Error probability — models how confusing a single screen is (bad copy, hidden button, unclear icon). Raising it increases both backtracks and abandonment.
- Screens in flow — a longer onboarding compounds the per-screen error probability across more steps, so completion rate drops faster than error probability alone would suggest.
- Batch size — more simulated users means the completion-rate estimate converges closer to its true underlying value (law of large numbers) instead of being noisy.