Ask four things.
Do not manufacture an answer.
At preregistered randomized intervals, participants answer one tiny instrument: keep, value, burden, and trust. Every due prompt remains in the denominator, nonresponse stays visible, and the result never enters W.
Tiny by design.
Exact by necessity.
Changing the wording, scales, timing, or response rules after seeing results breaks comparability. All four answers travel together as one prompt set.
Has the robot made your life better or worse recently?
How much work is the robot creating for you?
How comfortable are you letting the robot operate without supervision?
Random moments.
Fixed analysis.
The schedule is frozen before hour one. Delivery can drift by at most 24 hours, participants have 72 hours to respond, and one neutral reminder is the maximum.
Keep every prompt.
Rebuild every phase.
The checker validates schedule completeness, neutral collection, exact scales, explicit nonresponse, evidence bindings, and all overall and phase-level summaries.
Determines W and public rank.
Explains the trajectory; never changes W.