Remaining-time estimates support waiting decisions better than completion percentage alone

What it is

Remaining-time estimation translates task state into how long a person must wait or when completion is expected. Percentage does not reveal unequal later stages, queues, or external resources, while decisions to leave, switch work, or intervene depend directly on time.

Why it happens

Unequal stages make work fraction nonlinear in wall-clock time. A model using stage, historical distributions, load, and resource waiting can predict decision-relevant time and update with evidence. Frozen estimates followed by abrupt jumps destroy credibility; narrowing ranges reflect evidence better.

Studying it

Percentage, point estimate, interval, and stage-plus-time can be compared on wait decisions, timely return, intervention, trust, and prediction error. Variable-duration tasks require time-indexed absolute error and interval coverage. Interval comprehension is a separate outcome.

Where it stops holding

Exploratory tasks may only support an order of magnitude or “unknown.” A robot cannot promise external resources it does not control. Frequent updates distract during seconds-long tasks, and safety events need state and action before countdown.

Applying it

• Calibrate by task stage using real runtime distributions, showing a range and last update.
• Explain queue, retry, or dependency changes rather than animating an old number toward a new one.
• Monitor error, interval coverage, timely return, and repeated checks caused by bad estimates, stratified by task type.

Related

• Same group: X3.05.1 Long tasks need visible progress rather than silent execution · X3.05.3 Stalled progress should explain why rather than resemble failure · X3.05.4 Progress precision must match task predictability; false precision misleads
• Adjacent: X3.08 Communicating uncertainty · X4.05 Takeover and handoff
• Search terms: time remaining · completion-time prediction · prediction interval