Figure 4
Median deviation of each play from the track's own settled mid-life baseline, in percent of raw tune-out rate, at a track's first 20 plays (forward) and, mirrored, at its last 20 plays (backward). If a raised first play reflected genuine listener behavior at entry, the backward mirror would sit flat at zero. Removing the ambient turnover level from each play flattens the forward curve instead.
Forward, raw Backward mirror, raw Forward, ambient-corrected. Source: forward-versus-backward mirror test on raw play data, two stations; 90% bootstrap CIs in the text.