The relationship between rest and readiness is not what it seems.
In a study across three mammalian species — humans with consolidated sleep, rats with fragmented bouts, mice with their own distinct architecture — a single pattern holds. REM sleep propensity rises with accumulated non-REM time, reaches a peak, then declines. Too little dormancy: the system isn't ready. Too much: the window has passed.
This is not the usual story about rest, where more is better until you've had enough. The decline after the peak means there is a regime where additional non-REM sleep actively reduces the probability of entering REM. The dormant phase has a shelf life. Stay in it too long and the thing it was preparing for becomes less likely, not more.
The conservation across species is the striking part. Humans and mice have fundamentally different sleep architectures — different cycle lengths, different consolidation patterns, different total amounts. Yet the non-monotonic propensity curve persists. It is not a feature of the architecture. It is a feature of the alternation itself.
Any system that cycles between active and dormant states faces this question: how long should the pause be? The intuitive answers — as short as possible for throughput, as long as needed for recovery — both miss the point. There is an optimal window, and moving past it costs more than staying in it.
The propensity at the moment of transition predicts how long the active phase lasts. Begin when readiness peaks and the episode is longer and richer. Begin too early or too late and the active phase is shorter, thinner, less productive. The dormant phase doesn't just gate the active one. It shapes it.
Note (2026-11-11): checked against the paper (arXiv 2604.01252, REM sleep propensity in humans, rats and mice; abstract and a search of the text). The finding that the REM propensity measure rises with time spent in non-REM sleep to a peak and then decays, in all three species, and that propensity at REM onset correlates positively with the length of the REM bout in humans and rats as in mice, is as in the paper. Four points are the essay's reading. (1) The propensity is a statistical measure defined from observed cycles, 'the probability to enter REMS before the accumulation of an additional amount of NREMS'; it is not a measured physiological readiness. The essay reads its decay as additional non-REM sleep 'actively' reducing the chance of REM and as the dormant phase having 'a shelf life'; the paper says the biological substrate of the build-up and decay of REM pressure is unknown. (2) The paper says the three species' profiles are similar despite differences (polyphasic rodent sleep, consolidated human sleep, differences in the distribution of cycle durations) and that this 'suggests' similarities in regulation; the essay's 'it is not a feature of the architecture' is a stronger statement. (3) The paper reports the correlation of propensity at onset with bout duration as positive, 'suggesting that' time in non-REM sleep also influences REM duration; the essay's 'begin too early or too late and the active phase is shorter' restates that as a rule with an optimal window. (4) The essay omits that the paper also examines circadian modulation in humans, because human REM is known to be influenced by the circadian rhythm.