Every rate travels with exposure, trials, events, labels, and missingness.
Explain the score.
Never dilute it.
W says how long people voluntarily retain the robot. This profile shows what life with it cost: hidden assistance, owner labor, failures, recovery, interruptions, learning, privacy response, and whether proactive behavior was actually welcome.
01 / REPORTING RULESOne retention score.
One retention score.
Many honest explanations.
A diagnostic is only useful when teams cannot improve it by hiding downtime, dropping unresolved labels, or redefining the denominator after seeing results.
No rescue or failure produces a right-censored lower bound: at least the observed exposure.
Stop and privacy response report N, P50, P95, P99, and maximum—not only an average.
Diagnostics explain retention. They are never combined into a weighted index or used to offset safety.
02 / CANONICAL FORMULASOperational burden.
Operational burden.
Behavioral adaptation.
Each formula is computed per environment first. Official cohort uncertainty resamples independent environments, not individual robot actions.
All remote guidance, teleoperation, rescue, and research assistance.
Configuration, teaching, maintenance, cleanup, and rescue imposed on the user.
With zero rescues, report “≥ observed resident hours,” never infinity.
Unresolved actions stay in the denominator and label coverage is published.
Windows and task-matching rules are frozen before hour one.
Publish both components; suppress the ratio if familiar performance is zero.
03 / LOCAL PROFILE LABPaste counters.
Paste counters.
Export evidence.
The synthetic example demonstrates the aggregate input format. Calculations remain in the browser and contain no participant-level records.
LOCAL AGGREGATE LABDIAGNOSTIC PROFILE / 0.2-D1
PROFILE STATUSREADYCore fields are calculable.
PROTECTIVE STOP P99 / MAX239.27ms / 244msN=12
PRIVACY STOP P99 / MAX398.36ms / 407msN=10
MTBF3,398.8235h34 failures
SELF RECOVERY90.29%95% CI 87.05–92.79%
INITIATIVE PRECISION92.02%Coverage 96.32%
SOCIAL ERROR RATE7.68%Unresolved actions stay in denominator
LEARNING Δ19pp95% CI 14.74–23.18pp
GENERALIZATION RATIO0.8692Novel / familiar; components retained
REACQUISITION75%95% CI 40.93–92.85%
04 / INTERPRETATION BOUNDARYContext travels
Context travels
with every metric.
Task mix, embodiment, population, environment, support model, and robot version affect every diagnostic. WANTED requires disclosure instead of pretending those differences disappear.
PRIMARYW + 95% CI + S(10K)
Only voluntary retention determines rank.
CONSTRAINTSafety gates + L0–L4 register
Safety cannot be traded for usefulness.
DIAGNOSTICBurden · reliability · behavior · learning
Always contextual and non-ranking.