Sleep cycles, explained.

Wire a healthy 30-year-old to a polysomnograph, give them an eight-hour night, and score the recording in 30-second slices. What comes back isn’t a smooth slide into unconsciousness and a smooth climb out. It’s a staircase — down three flights, up two, down again, over and over until morning. Sleep scientists call that chart a hypnogram, and it’s the single most useful thing to have in your head when you’re deciding what time to go to bed.

The staircase has a rhythm, and the rhythm has a name: the NREM-REM cycle. You’ve probably heard it runs 90 minutes. That number is close enough to be useful and wrong enough to be worth correcting.

The four stages, and what each one is doing

Modern scoring, standardized by the American Academy of Sleep Medicine, splits sleep into three non-REM stages plus REM.

N1 is the doorway. Muscle tone starts to go and you get the occasional hypnic jerk — that falling sensation that snaps you awake. It usually lasts one to five minutes and accounts for 2 to 5 percent of the night. Wake someone from N1 and they’ll often insist they weren’t asleep. They were.

N2 is where you actually spend your life — roughly 45 to 55 percent of total sleep, more than every other stage combined. Its EEG signature is two odd little events: sleep spindles and K-complexes, both tied to memory consolidation and to keeping you asleep through noise.

N3 is slow-wave sleep, the deep stuff, 15 to 20 percent of a typical adult night. It’s the hardest stage to escape from — in lab settings some sleepers won’t rouse to sounds above 100 decibels. It’s also when growth hormone pulses and tissue repair runs hottest.

REM is the strange one, 20 to 25 percent of the night. Your EEG looks close to waking and brain metabolism rises by as much as 20 percent over quiet sleep — while your skeletal muscles are paralyzed, everything except the eyes and the diaphragm. This is where the vivid, narrative dreams live.

A cycle runs roughly N1 → N2 → N3 → back up through N2 → REM, then repeats.

A cycle is not 90 minutes

The best modern data comes from the Centre for Chronobiology in Basel, which went back through 26 years of lab recordings — 369 people, 1,567 nights, 6,064 scored sleep cycles, everyone given a controlled 8-hour sleep opportunity.

The median cycle came in at 96 minutes. The mean was 99.5. And the 95 percent interval ran from 60 to 150 minutes.

The number of cycles is also smaller than the textbooks suggest. “Four to five cycles a night” is the line you’ll read everywhere; in Basel’s eight-hour nights, most people completed three or four. One pattern is dependable, though: the first cycle is the shortest, mostly because the first REM episode is stunted — often under 10 minutes. REM episodes then get progressively longer all night, and the last one can run past an hour.

The night has two very different halves

Deep sleep is front-loaded. REM is back-loaded. That single fact explains most of what you notice about your own sleep.

N3 clusters in the first third of the night, in big 20-to-40-minute slabs during cycles one and two. By cycle three it’s thinned out; by cycle four it’s often gone entirely, and the back half of the night is essentially N2 alternating with long REM periods.

The practical consequences are asymmetric. Go to bed two hours late but wake at your normal time and you’ve mostly cut REM. Get dragged out of bed two hours early and you’ve kept your deep sleep and lost the same REM. Either way it’s REM that pays. This is also why you remember dreams on the mornings you sleep in — you’re waking out of the longest REM episode of the night rather than being yanked out of N2 at 6 a.m.

What waking at the wrong moment actually costs

The grogginess has a name: sleep inertia. In a 2006 JAMA study, subjects doing an addition task immediately after waking performed worse than the same subjects did after 26 hours of continuous wakefulness. The steepest impairment clears in the first 30 minutes, but full recovery can take up to a couple of hours.

Now the honest part. Sleep inertia is reliably worse when you wake during your circadian night, and reliably worse when you’re already sleep-deprived. Waking out of deep N3 is the factor everyone repeats — and the one the evidence is muddiest about. One study found a 41 percent performance drop waking from slow-wave sleep versus stage 2; other work found no stage effect at habitual wake times. Treat it as plausible, not proven.

So should you plan your alarm around cycles?

Yes, with the right expectations. The logic holds: waking out of light N2 generally beats being hauled out of N3, and since deep sleep is front-loaded, a too-short night is far more likely to end mid-N3 than a full one. Cycle-based planning — which is what our sleep calculator does — nudges your wake time toward the shallow part of the staircase.

What it can’t do is hit a specific minute. If your personal cycle runs 105 minutes and you’re planning in 90-minute blocks, you’re off by an hour after four cycles. Treat 90 minutes as a planning grid, not a guarantee — it’s why Hespera gives you a wake window rather than one alarm minute — and treat total sleep duration as the thing that matters most.

What bends the architecture

Age does the most. Total sleep time falls about 10 minutes per decade through adulthood; wake after sleep onset climbs roughly 10 minutes per decade between 30 and 60; REM’s share drops about 0.6 percent per decade.

Alcohol does something sneaky. A nightcap shortens sleep latency and consolidates the first half of the night, then wrecks the second half — more fragmentation, more awakenings, less REM at moderate and high doses.

SSRIs are the clearest pharmacological effect in the literature: reduced total REM and delayed REM onset, strongest early in treatment. If you’re on one and your dream recall dropped off a cliff, that’s why — and it’s not a reason to change anything without talking to whoever prescribed it.