← the guide

When they wake at two in the morning

0-8w · 2-4m · 4-7m · 7-12m · 12-24m · 2y+ · Updated 2026-08-25

A waking is not a broken night. It is the visible part of something that was happening all night anyway: a child surfaces between sleep cycles every twenty to forty-five minutes, and most of those surfacings you never hear about. Tonight you heard about one. That is the whole of what has been established so far — everything else is what you decide to do next, and almost none of it needs deciding at two in the morning.

Nights have seams

An adult sleep cycle runs ninety to a hundred and ten minutes. A baby's runs about fifty. Between cycles there is an arousal — a partial surfacing that is part of the architecture, not a fault in it — and babies naturally wake two to eight times a night on their way through.

Two structural facts make the seams uneven, and both are worth knowing because they change what a waking at a given hour means:

So a waking at half past ten and a waking at half past four are not the same event, and it is the four-thirty one that is closest to ordinary physics. In pooled reference data the longest single sleep period runs about 5.7 hours under six months and about 8.3 hours from six to twenty-four — those are the long stretches, in the studies everyone quotes.

How many is normal — and why the published number will make you feel like a failure

Here is the thing I most want to put in front of a parent at 2 a.m.

The reference figure in the literature, pooled across dozens of studies, is a mean of about 0.8 night wakings per night across infancy. Set against my own logs — 670 nights of one child, each waking recorded by my partner or me as it happened, on the review's own age bands:

In the newborn band the two roughly agree. After it, mine run about three times higher, in every band.

I do not think my child is unusual. I think the two numbers are measuring different things. Most of the pooled studies ask parents, retrospectively, how many times their baby wakes at night — and parents are known to under-report it, because wakings a child resolves alone are invisible and because recall flattens a month of nights into an impression of one. Night waking is already the shakiest number in that review, the one its own authors report the least agreement on. Log each waking as it happens and you find more of them. Of course you do.

My guess at why the gap opens after the newborn months — and it is a guess, not a measurement — is that early wakings are memorable because each one is a feed and you are up anyway, while later ones are brief, blurry and easy to lose by morning. The under-counting gets worse exactly as the wakings get cheaper.

The practical consequence is unpleasant and easy to fix. A parent who starts tracking and then compares their real log against the reference "0.8" will conclude that something is wrong with their child. Nothing is wrong. They are comparing a measurement to a memory. Compare logs to logs.

The peak is not where you expect it

The same data says something I did not expect and would have wanted to know in advance: the worst stretch was not the newborn months.

Wakings peaked at seven to eleven months — a mean of 3.5 a night, and as many as eight — well above the 2.2 of the first two months. Then they fell back.

That shape is not mysterious once you line it up against what else is happening at that age. Separation anxiety begins around six months and peaks at eight to nine. The eight-to-ten-month regression lands there, lasting one to two weeks and affecting something like forty to sixty per cent of babies. New motor skills — rolling, sitting, pulling to stand — reliably wreck sleep for days at a time until the skill is mastered, and that is exactly the season for them. And the third nap is being dropped somewhere in the same window.

If you are holding a four-month-old and someone has told you the hard part is nearly over, the honest version is: the most broken nights of my two years were still ahead. It sounds like a bleak thing to publish. I found it enormously steadying at the time, because a night that is expected to be bad is a different experience from a night that feels like evidence of a mistake.

The second half of the shape is the encouraging half. As wakings became more frequent they also became shorter — a median of 27 minutes awake per waking through the first six months, 20 minutes at seven to eleven, 14 minutes in the second year — and nights with no waking at all went from none, to 2%, to 7%, to 18%. The night does not resolve by the wakings stopping. It resolves by them getting cheap.

(One child, hand-logged, and a waking we slept through is a waking we did not record. The trend held across every band, which is more than noise usually manages.) Both of these results are written up with their full workings in the notebook — the measurement gap and the seven-to-eleven-month peak.

What actually changes a night

Almost everything that improves nights is a daytime lever, which is the frustrating and liberating part.

In the moment itself, both books say the same short thing: keep it dim, keep it boring, keep activity low. A night waking that is handled like a small daytime event tends to become one.

What NaniNani does with it

Night waking is a state of its own here, with two exits, and the distinction is real rather than cosmetic: back to sleep continues the same night, while up for the day starts the morning and begins a new day's pressure curve. Choosing the wrong one is the single most consequential mislog in the app, which is why they are two separate buttons and not a guess.

While a waking is open, the app treats it as awake time and lets pressure rise, because that is what is physically happening. It does not predict wakings, and it will not pretend to — a two-process model describes how sleep pressure moves, not when a tooth hurts or when a nine-month-old remembers you are in a different room.

And it does not count them against you. There is no waking tally on the home screen, no streak that a bad night breaks, no nightly grade. The number in this article exists so you can see the shape of a normal year; it is deliberately not a number the app shows you at 2 a.m., because at 2 a.m. it would stop being information and start being a verdict.

Where this comes from