A range reflects uncertainty more honestly than a single precise number

What it is

Extrapolation yields a point, and around the point is a band: rate may still change. Collapsing the band into “1:17 left” spends second-precision on minute-scale uncertainty. A range (“about 1–3 minutes”, “under five minutes”) writes the uncertainty into the shape of the number, so people allocate attention against a band rather than against a point the next second’s formula will move. Honesty here is a representation, not an apology afterwards.

Why it happens

A point estimate forces a plan it cannot fund: I will come back in 1 minute 17 seconds. The plan’s grain is seconds; the noise that produced the point is coarser. When the noise shows, the point has to move house, and on screen that is a changed number. A range speaks at the noise’s grain from the start: if the band is wide enough, rate jitter inside it can leave the copy still; only a band breach changes it. One fewer change is one fewer “the number moved” event. Extrapolation’s jump is absorbed in the width instead of being displayed as a new point on the next frame.

A range also encodes “the system knows it is uncertain” in the form. A point, however conservative the algorithm, looks like a claim to precision. The claim itself raises the scoring bar: twenty seconds off counts as missing. A range rescores to “landed in the band”, aligned with the method’s real resolution.

Where it stops holding

On a negotiated pipe with homogeneous leftover work and a short remainder, point-estimate noise is under a second and a range looks affected (“about 0–2 seconds”). A point, or the bar alone, is cleaner. Safety deadlines and OTP countdowns need a deterministic cutoff, not an estimate band; this leaf must not turn those into ranges. A band drawn too wide (“about 1 minute to 1 hour”) has about as much information as no number; people ignore the range and watch the bar. Too wide is not saying. The ends of the band must come from dispersion in the rate window, not from a casually written “a few minutes” as tone.

Applying it

• When variation in the rate window crosses a threshold, show a range or an upper bound (“under 5 minutes”), not a second-precise countdown.
• Let width follow the window: noisier window, wider band; after it settles, the band may narrow. Narrowing is a revision, not a change of story.
• Do not stack a range and a point (“1–3 minutes (1:17 left)”). The point eats the range’s honesty.
• How to check: watch copy on a jittery network. A point estimate will change every second; after switching to a range, if completion still lands inside the last published band, the representation fits the method’s resolution better than a point. If the band is never breached and never narrows, it was drawn with no information.

Related

• Same group: I2.10.1 Remaining-time estimates extrapolate recent rate; a rate shock makes them jump · I2.10.3 Revise the estimate as progress continues; do not freeze the first figure · I2.10.4 With no sound basis, show elapsed time rather than invent remaining
• Nearby: I2.02 Predictable progress · I4.06 Countdowns and deadlines · E6.09 Determinate and indeterminate progress
• Search terms: interval estimate · uncertain ETA · range remaining time