Fitness associated with a given arrival date at a breeding site varies in three different years. The peak in each year (A) denotes the optimal arrival date, T*, and we assume that when many years are considered T* is normally distributed (shown along the y-axis in panels B through D). Bottom: B) The annual optimal arrival date may be predicted by a cue at a stopover or wintering site, with the between-year distribution of the cue shown as a normal distribution along the x-axis. The regression line shows the expected optimal arrival date for a given cue value, and the envelope indicates the uncertainty of the cue. C) Regressions on the cue can vary: in this case the cue provides more precise information (lower variance). D) If the distribution of the cue is narrower, there might be a larger cost of assessing the cue wrongly, and one could expect selection on individuals to estimate the cue better through enhanced perception, cognition, or increased sampling.