For about six months in 2022 I went to bed by a calculator. The math, if you have not encountered it, goes like this: a sleep cycle is ninety minutes, so a full night is some round multiple of ninety minutes — six hours, seven and a half, nine. Wake up at the end of a cycle and you feel rested; wake up in the middle and you feel as though someone has dropped a piano on your chest. The promise is that you can shave thirty or forty-five minutes off your night and feel better, because what you cut is not sleep but the wrong kind of sleep. I went to bed at 10:42 p.m. so I could wake up at 5:12 a.m. and feel, by arithmetic, refreshed.
It did not work. Some mornings I felt fine; some mornings I felt like I had been hit by a small truck; one Tuesday in March I slept through two alarms and the dog. After enough of this I went back to reading the sleep literature, which I should have done before I started doing math.
Where the number comes from
The ninety-minute figure is real, in roughly the same way that “the average American household has 2.5 people” is real. It is a population average, drawn largely from polysomnography studies done in sleep laboratories in the 1960s and 70s. In those studies, researchers tracked the cyclical movement of sleepers through the four stages now sometimes called N1, N2, N3, and REM, and they noted that one full pass through those stages took, on average, somewhere around ninety minutes.
The trouble is the word average. The range in healthy adults runs from roughly seventy to a hundred and twenty minutes, and individual variance from one cycle to the next, within the same night, can be substantial. The first cycle of the night tends to be shorter and N3-heavy; later cycles get longer and more REM-dominant. If you set your alarm to a strict ninety-minute schedule, you are assuming a metronome where the body keeps a more honest rhythm — close, but not on the click.
What sleep researchers actually look at is sleep architecture: the shape of how the night moves through its stages. A well-rested brain spends the first third of the night largely in deep, slow-wave sleep (N3), where the most pronounced restorative work appears to happen, and the last third of the night largely in REM, where memory consolidation and emotional processing seem concentrated. The middle of the night is a busy in-between. None of that is on the clock face of your phone.
What it actually feels like to wake up well
If you have ever woken up just before your alarm and felt, briefly, like the kind of person who could probably learn Italian, you have experienced what the literature calls a low-arousal-threshold awakening. You came up out of REM or light N2 sleep, where the body is closer to the surface; the transition was short, and the so-called sleep inertia — the grogginess of a brain still flushing adenosine — cleared in a minute or two.
Being yanked out of N3 produces the opposite. Sleep inertia after deep slow-wave sleep can last twenty to forty minutes, and during that window your reaction time, mood, and decision-making are demonstrably worse than they will be by 9 a.m. This is the sensation the ninety-minute calculator is trying to help you avoid, and the underlying instinct is reasonable. Waking up out of deep sleep is unpleasant. It is the implementation that is wrong.
What sleep apps are actually measuring
The popular sleep tracking apps and smart alarms do not, generally, watch your brain. They watch your wrist, or your mattress, or your phone’s microphone. From that data they infer a stage — usually by guessing that movement means light sleep and stillness means deep sleep — and they set the alarm to fire during a window when you appear to be drifting upward. Compared to lab polysomnography, the agreement is mediocre. Validation studies on consumer wearables have generally found high accuracy at detecting sleep versus wake and considerably worse accuracy at distinguishing N3 from REM, which is the distinction that actually matters for the question of when to wake you up.
This is not to say the apps are useless. A device that nudges you to be in bed by 10:30 most nights is doing meaningful work, even if its cycle detection is more art than science. The misuse is in trusting its readouts as precision instruments. You are not getting a sleep stage report. You are getting a guess from a watch.
What to do instead
The variables that the strongest sleep research keeps returning to are unglamorous. Total sleep duration matters — for most adults, seven to nine hours in bed. Timing consistency matters; people who go to bed and wake up within roughly a thirty-minute window each day appear to sleep more deeply than people whose schedule wanders, even when the average duration is the same. Light exposure in the first hour after waking matters. Caffeine after noon matters. The hours before sleep matter more than the minutes inside it.
If you want a practical handle on the ninety-minute problem, here is the version I have come around to. Decide what time you need to wake up. Count back eight hours from there to get a bedtime window — not a target. Allow yourself to fall asleep anywhere inside a fifteen- or twenty-minute span. Avoid setting an alarm so tight that you are panicked about a four-minute deviation, because the anxiety of overshooting the window will cost you more sleep than the cycle math saves.
The honest answer is that most of us do not know, in real time, what stage of sleep we are in, and we cannot — with current consumer tools — reliably wake ourselves out of light sleep on purpose. What we can do is sleep enough, sleep on schedule, and let the architecture take care of itself. The cycles, on average, do their work. You just have to stop trying to outsmart them.
