Guides

How naido forecasts sleep

As of: 12.10.2026

naido forecasts your baby’s next sleep as a time window. The calculation runs entirely on your phone and uses only your child’s entries. This page describes the method in naido 0.28.4, including its limits.

What naido measures: wake windows

A wake window is the time from waking up to falling asleep again. naido reads the wake windows from your entries. Short interruptions count as a pause within the same sleep: up to 15 minutes for naps, up to 30 minutes at night. Wake windows under 15 minutes or over seven hours are left out as a likely gap or typo.

Similar situations count more

Once there are at least ten comparable wake windows, naido takes for the forecast the wake windows of the last 120 days at most (older ones only if there are none in that period) and weights each one by how similar the situation back then was to today’s:

• the time of waking up
• which nap of the day is next
• how long the sleep before it was

On top of that, recent days count more: a wake window from a week ago counts half as much as one from today. The middle of the window is the weighted median, the value with half of the weight below it and half above.

At the start, with fewer entries, naido only takes the wake windows before the same nap of the day, counts recent ones more and mixes them with age-typical starting values from published reference tables. The more you log, the more your own data counts; from ten matching observations on, naido uses only them. Importing from another app shortens this run-up.

The night

For the night, naido blends the calculation half and half with your usual bedtime, because bedtime is a fixed habit in many families. Waking up in the morning is forecast directly from the usual wake-up time of the last nights.

A window instead of a minute

The width of the window follows your child’s variability: if wake windows vary a lot, it gets wider. With your own data it lies between ±12 and ±35 minutes around the middle, before the night up to ±50 minutes; with no entries at all, naido uses the range from the age table. In disruption mode the window can become up to 1.6 times wider, so up to about ±56 minutes, before the night up to about ±80. In addition, naido shows in three steps how solid the forecast is. In the settings you can choose a fixed window of ±20 or ±30 minutes instead of the adaptive width.

Measured on real data, the adaptive, narrow window is right in about two out of three cases. The fixed ±30-minute window is an hour wide and is right more often, but says less. You choose what helps you more.

“Getting tired”

When you tap “Getting tired”, naido sets a short window around ten minutes after the tap, ±12 minutes. The signal stays valid for 45 minutes. In simulations on real data, such a signal hit far more often than any model built from log data, because it adds new information: what you observe right now.

Exceptions

Teething, illness, sleep regression: in disruption mode the most recent days count much more (with little data the window gets wider instead), and naido rates the forecast as less solid. Periods like a holiday can be marked, even afterwards; during them naido uses the most recent days, and forecast quality leaves those days out.

Check the hits yourself

Under Statistics, forecast quality shows how often the forecast would have hit for your child. naido recalculates the sleeps in the selected period (7, 14 or 30 days, 14 by default), each using only the data that existed at the time. A sleep only counts once there were at least ten comparable wake windows before it, so if you start without any history, the first days are not counted yet. Imported entries count as history, so after importing enough data the rate counts right away. Next to the hit rate it shows the median deviation, because very wide windows would make any rate easy to reach. Groups with fewer than five cases show no rate.

What naido deliberately does not do

We tested more elaborate methods on nine months of real data: machine learning such as gradient boosting and random forests, regression, and the two-process model from sleep research. None hit measurably better than the weighted estimate from the last few days. A baby changes faster than a model can learn.

Limits

• Even a method that knows each day’s average in advance hit only about half of the naps to within ±20 minutes in our measurements. The wake window varies by about half an hour from day to day. That is why naido shows a window and not a minute.
• The measurements rest on few real histories: the longest over nine months, some parts checked on four children. Your child may differ; forecast quality shows how it goes for you.
• naido observes and calculates. The app gives no medical advice; if you are worried about your child’s sleep, talk to a doctor.