A sharp transition 1 °C wide lives in the sample. Temperature jitter contributes to feature separability in this simplified measurement model — detector noise, signal amplitude, sampling, uniformity, fitting, calibration and replication all matter too.
Schematic step (sigmoid, 1 °C width) convolved with Gaussian control noise of the chosen σ; the noisy trace resamples on each move. Main curve = the Gaussian-weighted expected response; faint dots = one random measurement pass. A fluctuation band materially narrower than the feature is generally desirable, but no fixed ratio guarantees resolution. A datasheet stability band is not automatically a one-standard-deviation Gaussian, and separability here refers to this simplified signal model only — not to any real stage’s specification.