What a sleep score actually measures — and what it cannot
Every sleep app hands you a number out of 100. Very few will tell you what went into it. Here is one score taken apart completely — the inputs, the weights, the grade boundaries, the things that were removed and why — and an honest account of what a number built that way can and cannot mean.
You open a sleep app and it tells you last week was a 78. Or a B. Or that your "sleep quality index" is 6.4 out of 10. The number has a colour, and the colour has an opinion about you.
Almost none of these apps will tell you what the number is made of. That is a problem, because a score is not a measurement — it is a formula somebody chose, and until you know the formula you cannot tell whether a 78 means you slept badly or means the formula dislikes the way you live. So here is one taken apart completely, including the bits that are unflattering.
The whole formula
SleepMind’s weekly score is built from exactly two things, out of 100:
- Duration — 55 points. Your average nightly hours across the window, divided by a seven-hour target, capped at 1. Seven hours gets all 55; three and a half gets about 27; ten hours gets 55, not more.
- Quality — 45 points. Your average self-rating across the window, out of five, times 45. An average of 4 gets 36.
The two are added. The grade boundaries are 85 for an A, 70 for a B, 55 for a C, and below that a D. Every chart on the page covers the last seven days rather than your whole history, so the number keeps moving once you have months of data behind it rather than going dead exactly when you have earned the right to watch it change.
That is the entire mechanism. There is nothing else in it. You can work out your own score on paper, and if you do not like it, the honest response is to disagree with the weighting rather than to wonder what the app knows.
Why there used to be a third input, and why it is gone
The score originally had three components, the third being mood, worth 25 points. It was removed, and the reason is worth telling because it illustrates how a number can be quietly broken while looking perfectly healthy.
Nothing in the app ever asked for mood. There was no input for it anywhere, so every record sat at the schema default of 3 out of 5 — which meant 15 of those 25 points were awarded, automatically, to everybody, forever. A quarter of the scale was doing nothing except flattening the other three quarters. It also capped the achievable score at 90 while an A needed 85, which made the top grade far harder to reach than the stated boundary implied.
The fix was to delete the component and redistribute its weight. The general lesson is worth carrying to any score you are shown: a component that cannot vary is not a component, it is a constant, and a constant in a score is just a thumb on the scale.
The target and the chart used to disagree
A second bug, same family. The score divided by eight hours while the charts drew their "on target" line at seven. A seven-hour night was therefore painted green as a good night while quietly earning 48 of 55 available points. A week of hitting the goal the interface displayed produced a grade the user had no way to explain from anything on screen.
There is now one number, exported from one place and read by both. This is the mundane kind of defect that produces most unfair-feeling scores in most apps: not malice, just two copies of the same fact drifting apart.
Which suggests a reasonable question to ask of any sleep score you are given: is the target it grades you against the same number as the target it displays to you? Surprisingly often, nobody has checked.
What this score genuinely cannot tell you
Now the important half. There are things this number is structurally incapable of knowing, and no amount of careful weighting fixes them.
It does not know whether you slept. It knows what you told it. SleepMind has no microphone access, no motion sensor, no wearable connection — every hour in it was typed by you, from memory, the next morning. That is a reasonable basis for a journal and it is not a measurement of anything.
It does not know about sleep stages. It cannot. Nothing on a phone screen can tell you how much REM or deep sleep you had, and any app that puts a stage breakdown in front of you on the basis of self-report is inventing it. Even wrist-worn devices infer stages from movement and heart rate and disagree with clinical measurement more than their interfaces suggest.
It does not know if anything is wrong. It cannot detect apnoea, insomnia, restless legs, or anything else. A run of Ds means you have been reporting short or poorly-rated nights. Only a clinician can say what that is, and if the pattern is persistent or you feel wrecked during the day, that is the conversation to be having rather than a scoring question.
It does not know your life. Seven hours is a reasonable general anchor, not a personal prescription. It is not adjusted for your age, whether you have a newborn, whether you are on a rota, or whether you happen to do well on six. A rotating shift week will grade badly no matter how skilfully you handled it — which is a fact about the formula, not about you.
And it does not know how the day went, which is the thing you actually care about. The score is a description of a record, not a verdict on a week.
What it is genuinely useful for
Read as a trend line rather than a grade, a simple score does real work.
It is consistent. The same inputs always produce the same output, so a fall from 80 to 65 reflects something that changed in what you reported. That is exactly what you want a trend for.
It is decomposable. The page shows the duration and quality contributions separately, so "my score dropped" always resolves into either "I slept less" or "I rated it worse", and those have completely different follow-up questions.
It is short-window. Seven days means a bad fortnight shows up and then clears, instead of being drowned by a lifetime average. A metric you cannot move is a metric you stop reading.
And alongside it sit the things that are not scored at all: how many nights carried a dream, which feelings keep coming back, and what you tagged as getting in the way — noise, heat, cold, stress, the phone, stuffy air, light, a late bedtime. Those eight tags are frequently more actionable than the number, because each one names something you could in principle change tonight.
How to read your own score without it reading you
- Ignore the absolute value. The first useful number is your fourth week, compared with your third.
- When it moves, look at which of the two bars moved. Duration and quality are different problems.
- Check the tags before you change anything. Three weeks of "phone" is a more specific instruction than any score.
- Discount weeks the formula was never built for — a rota, a newborn, a flight, an illness. You know those weeks were unusual; the arithmetic does not.
- If the number is making you anxious about sleeping, stop looking at it. That failure mode is real, it is well recognised among people who track their own sleep, and worrying about a score is not a route to sleeping better.
- If the trend is genuinely bad and you feel bad, take the record to a doctor and stop asking an app.
A score is a summary of what you wrote down. Treated as that, it is a useful, cheap, honest instrument. Treated as a grade for how well you performed at being unconscious, it is worse than nothing.
Frequently asked questions
Why is the duration target seven hours and not eight?
Because seven is the figure the interface already draws its "on target" line at, and grading against a stricter number than the one you display is the taskmaster failure — it hands people a score they cannot explain from anything on screen. Seven is an anchor for a general adult, not a prescription for you; some people do well on six and some need nine, and a formula cannot tell which you are.
Can I get above 100, or lose points for sleeping too long?
No to both. The duration component is capped at the target, so ten hours scores exactly what seven does. Nothing in the score penalises a long night. If you want to know whether long nights are helping you, the quality rating you gave them is a better guide than the score is.
Why does my score not include sleep stages, like other apps?
Because the app has no way to know them. It sees what you typed, and no amount of processing turns a bedtime you remembered into a REM figure. Any app showing you a stage breakdown derived from self-report is showing you a fiction, and even wearables infer stages indirectly and disagree with clinical measurement more than their charts imply.
My score is a D every week. Should I be worried?
It means you have been logging short nights, poorly-rated nights, or both — that is all it can mean. Whether it is worth worrying about is a medical question and an app cannot answer it. If the pattern has lasted weeks and your days are suffering, see a doctor, and bring the record: "six hours or less on eighteen of the last twenty-one nights" is a much better opening than "I sleep badly".
Does the score include naps?
Yes — every stretch you log on a day counts towards that day’s total, naps included. Each stretch also keeps its own rating, so a good nap and a bad night stay distinguishable underneath the average rather than being blended into one middling number.
Write it down before it goes
SleepMind puts the dream note on the first screen, so you can tap a feeling and type a fragment before you have stood up. The sleep log is folded one tap below, on a 24-hour timeline that takes naps, split nights and late shifts. Free, on iPhone, with no sign-up wall.