Free tool · no signup
Sleep Regularity Index calculator
The Sleep Regularity Index (SRI) scores how similar your sleep is from one day to the next. It takes every minute of the day, asks how often you were in the same state (asleep or awake) at that minute on two consecutive days, and rescales that agreement so 100 means identical every day and 0 means no better than chance. The formula runs from -100 to 100; in a 60,977-person UK Biobank cohort the median was 81, with the middle half scoring between 74 and 86. Enter at least seven nights of bed and wake times and this page computes it in your browser.
Runs in your browser · Nothing uploaded · No account, no email
Sleep Regularity Index
in your browser
Day names are for orientation only. The score depends on the order of the nights, not on which weekday each one is. Up to 14 nights.
More regular than most
This week sits in the fourth quintile of the UK Biobank distribution. Most nights line up; one or two drift by an hour or more.
Avg sleep
7h 44m
Across 7 nights
Bedtime spread
±46m
Standard deviation
Wake spread
±66m
Standard deviation
Shortest
7h 20m
Your thinnest night
Night 5 (Fri) costs you the most. Move it to your median night, 23:25 to 07:00, and this week scores 90.5 instead of 84.3. Nothing else about the week changes.
One bar per night, bed to wake, on a single clock axis. The dashed lines are your median bed and wake times. Bars that line up are the minutes the index counts as matching.
The week above is an example, so the page has something to show. Overwrite it with your own nights. Everything is computed in this browser, nothing is uploaded, and nothing is saved, so reloading brings the example back.
Reading the result
What your score actually means
- 87.32 and up
Top fifth
This week is more regular than about 80% of adults measured by wrist accelerometer in UK Biobank. Your bed and wake times land within roughly the same hour every day.
- 83.16 to 87.32
More regular than most
This week sits in the fourth quintile of the UK Biobank distribution. Most nights line up; one or two drift by an hour or more.
- 78.56 to 83.16
About average
This is the middle fifth of adults, around the UK Biobank median of 81. A typical shape here is five steady weeknights and a weekend that moves.
- 71.65 to 78.56
Less regular than most
This week falls in the second-lowest quintile. Either several nights drift, or one night is far enough out of step to pull two comparisons down at once.
- below 71.65
Irregular
This week is in the least regular fifth of UK Biobank adults, the group with the highest mortality risk in that cohort. Your sleep timing changes enough that few minutes of the day are spent the same way two days running.
The published formula runs from -100 to 100, but a negative score means your sleep timing agreed with itself less often than chance, which takes a rotating shift pattern or a typo. The 61 undergraduates in the original 2017 study ranged from 38 to 86, mean 73.
What moves it
What moves the number, and what doesn’t
- 01
A weekend lie-in costs double
Take a week of seven identical nights, 23:00 to 07:00, which scores exactly 100. Let one night run to 10:00 instead. That single three-hour lie-in drops the week to 91.7, because it breaks the comparison with the day before and the day after, 180 minutes each. Put the same lie-in on the last night, where there is no day after, and the week still scores 95.8.
- 02
Bed time and wake time cost the same
Push one bedtime an hour later and the week goes from 100 to 97.2. Pull one wake time an hour earlier instead and it goes to 97.2 as well. The index has no preference between the two ends of the night. The reason to fix your wake time first is not the arithmetic, it is that you control an alarm and you do not control when you fall asleep.
- 03
It says nothing about how long you sleep
Sleep five hours a night at exactly the same five hours and you score 100. That is not a flaw. Lunsford-Avery and colleagues found irregularity predicted cardiometabolic risk independently of sleep duration, which is the argument for keeping regularity as its own number rather than folding it into a sleep score.
The calculator runs the same test on your own week. It swaps each night in turn for your median night and reports the one that buys the most, which is usually worth a handful of points rather than the twenty people expect. Irregularity is normally spread across a week, not parked in one night.
The method
The formula, and what this version cannot see
Sleep Regularity Index
SRI = -100 + (200 / (M·(N-1))) · Σj=1..N-1 Σi=1..M δ(si,j, si,j+1)
Split the day into M epochs and take N days. si,j is whether you were asleep or awake at epoch i on day j, and δ is 1 when two states match and 0 when they don’t. So the double sum counts the epochs that matched the same epoch 24 hours earlier, and the rescaling turns full agreement into 100 and no agreement into -100. This page uses M = 1440, one epoch per minute, and days that run noon to noon so a 01:20 bedtime belongs to the night before it rather than splitting across a midnight boundary.
Published SRI values are not computed from two times a day. They come from per-minute actigraphy or detailed diaries, which record naps, the twenty minutes you spent staring at the ceiling at 3am, and how long you took to fall asleep after the light went off. This page models each night as one unbroken block between the two times you typed. Every awakening it can’t see is a stretch of minutes that would have had a chance to disagree with the day before.
The direction of that error is predictable: this calculator scores higher than an accelerometer would for the same person, typically by a few points. It is still a fair comparison of one of your weeks against another, because the same simplification applies to every week. It is not a fair comparison against a number from a paper.
Two smaller things worth knowing. Seven nights gives six comparisons, so one unusual night is a sixth of the total and moves the score a long way. And because the score only looks at consecutive days, it has no concept of weekday or weekend. Shuffle your week into a different order and you get a different number.
Sources
- [1]Irregular sleep/wake patterns are associated with poorer academic performance and delayed circadian and sleep/wake timing
Phillips, Clerx, O’Brien, Sano, Barger, Picard, Lockley, Klerman & Czeisler, 2017 · Scientific Reports 7, 3216
The paper that defined the Sleep Regularity Index. 61 undergraduates kept sleep diaries for about 30 days each; their scores ran from 38 to 86, mean 73 ± 11. This page implements the formula exactly as printed there.
- [2]Validation of the Sleep Regularity Index in Older Adults and Associations with Cardiometabolic Risk
Lunsford-Avery, Engelhard, Navar & Kollins, 2018 · Scientific Reports 8, 14158
Took the index out of a student sample and into 1,978 older adults in the MESA cohort. Irregularity tracked cardiometabolic risk independently of how long people slept, which is the reason regularity is worth its own number.
- [3]Sleep regularity is a stronger predictor of mortality risk than sleep duration: a prospective cohort study
Windred, Burns, Lane, Saxena, Rutter, Cain & Phillips, 2024 · Sleep 47(1), zsad253
Scored 60,977 UK Biobank adults from wrist accelerometry. Median SRI 81.0 (IQR 73.8 to 86.3), and the quintile cut points from this study are the bands this calculator reads your score against.
This scores one week. The point is the fifty-second.
You typed this week in by hand, which is fine once. Habit Pocket logs bed and wake times as real times on iPhone and the web, and keeps this same candle chart current without you rebuilding it. A regularity score is only worth anything as a series.
5 habits free, forever. No credit card. Your data stays private.
FAQ
Sleep regularity, answered
In the largest study to measure it, 60,977 UK Biobank adults had a median SRI of 81, and a score above 87.3 put a person in the most regular fifth. Below 71.7 was the least regular fifth, the group with the highest mortality risk in that cohort. These are population percentiles rather than clinical thresholds, so there is no line you cross into “bad”. The useful comparison is your own score this week against your own score last month.
For mortality risk in one large cohort, it predicted better. Windred and colleagues scored 60,977 UK Biobank adults on both and found the most regular fifth had roughly 30% lower all-cause mortality risk than the least regular fifth, with regularity fitting the data better than duration did. That is one observational study, and it does not prove that making your sleep more regular makes you live longer. It does mean regularity is worth a number of its own, which is more than most sleep apps give it.
Social jetlag measures one gap: the difference between the midpoint of your sleep on work days and on free days. The SRI compares every minute of every day with the same minute the day before, so a Wednesday you could not fall asleep counts as much as a Saturday lie-in. Social jetlag is easier to work out on paper and it speaks directly to the weekday-weekend split. The SRI needs a full timeline and returns one number that does not care which day of the week it is.
The formula runs on two nights, because it compares each day with the day before, but two nights is one comparison and tells you almost nothing. The 2017 paper that defined the index used 26 to 36 days per person. Seven nights is the practical minimum for a number worth reading, since that is the shortest span containing a weekend, and this calculator accepts up to 14.
Because this page models each night as one unbroken block, from the bed time you typed to the wake time you typed. Published SRI values come from per-minute actigraphy or diaries that also capture naps, the twenty minutes you spent awake at 3am, and the time it took you to drop off, and each of those adds minutes that can disagree with the day before. Expect the number here to run a few points high against an actigraphy score. It is still a fair comparison of one of your own weeks against another, because the same simplification applies to both.
Last updated August 2026 · I implemented the formula from the 2017 paper myself and checked every worked example on this page by hand · By Bohdan Stefaniuk