The flagship measure

Sleep regularity, and why it beats sleep duration

Almost nobody knows their own, and it predicts more than the number everybody does know.

The Sleep Regularity Index asks one question: if I pick a random minute of the clock, how likely is it that you were in the same state, asleep or awake, at that minute yesterday and today? Rescaled, that probability runs from −100 to +100.

What the number means

• +100 is a perfectly repeated schedule: asleep and awake at exactly the same clock minutes every day.
• 0 is chance. Knowing what you were doing yesterday at 2am tells you nothing about today.
• Negative means actively antiphase, awake now at the hours you were asleep before. Shift work lands here.

Most people are surprised twice. First that the population median is as high as 81, and then that their own is lower than they assumed. A regular sleeper's fortnight, drawn as one row per day with sleep inked in, has a clean vertical band running down it. An irregular sleeper's is scattered. That image is the whole idea, and it is the picture Even's interface is built out of.

Why it beats duration

Sleep duration is the number everybody tracks and the one every app leads with. In the largest study of the two measures together, regularity was the stronger predictor of mortality risk, and it stayed the stronger one after duration was accounted for. Windred and colleagues published that in Sleep in 2024, from UK Biobank accelerometry.

The mechanism is not mysterious. Your body runs on a clock, and the clock is set by when things happen rather than by how much of them there is. Seven hours taken at a different time every night asks the clock to move every night. Seven hours taken at the same time does not. It is also, unlike duration, something a person with a fixed and demanding life can often actually change.

How Even computes yours

From the sleep already in Apple Health, so there is nothing to log and no wearable to buy beyond whatever you already sleep in. Even needs at least seven usable days with at least six adjacent pairs among them before it will show a value: a Sleep Regularity Index computed from four nights is not so much wrong as meaningless, and showing it would be the sort of invention this app is built to avoid.

One deliberate departure from the published formula

The canonical denominator counts every possible pair of minutes across the window, whether or not data exists for both sides. On research-grade accelerometry worn continuously that is exactly right. On an Apple Watch that spends some nights on a charger it is not: a missing minute scores as a mismatch, so the metric quietly punishes people for not wearing a watch and calls the result irregular sleep.

Even counts only the pairs where both minutes are actually known. A night of missing data therefore lowers the confidence Even reports, which is honest, instead of lowering your score, which would be a lie. It is the one place the app departs from a published formula, and it is written down here for the same reason it is written down in the source.

Raising it

Regularity is one of the few sleep measures that responds to a single decision rather than to a lifestyle overhaul, and the decision is about anchoring rather than about earliness.

• Hold the wake time first. It is the anchor the clock actually reads, and it is usually easier to hold than a bedtime, which depends on how the evening went.
• Shrink the weekend gap. An hour of drift on Saturday and Sunday costs more regularity than most weekday variation does, because it moves two days out of seven.
• Protect the shape, not the total. A slightly short night at your usual hours does less to your index than a long one taken three hours late.
• Give it a fortnight. The index is computed over a window, so a change shows up gradually rather than tomorrow morning.

Even shows your value against the population median as a ring with the median notched on it, so you can see the distance rather than being told a word for it. It also tracks it over time, which is the part that matters: the interesting question is not what your regularity is, but which direction it is going.

The source

Median 81.0 and an interquartile range of 73.8 to 86.3 are the two statistics Even claims from that paper, and they are all it claims. Quintile cut-points appear in the work but have not been verified against the source here, so Even has no quintile language and will not acquire any until they are.

Even describes patterns in your own data. It is not a medical device, it does not diagnose or treat anything, and nothing in it is medical advice. Persistent sleep problems are worth taking to a clinician.