How we measure
The phone sits on your bedside table or the floor, not the mattress, so the microphone leads. It hears how loud the room is, the rhythm of your breathing, the rustle of turning over, and what kind of sound it is. The motion sensor still has a job: it notices when you pick the phone up, snooze it or get out of bed.
Every minute the app writes down a loudness reading, how regular your breathing sounded, how much rustling it heard, and the seconds of any snoring, talking, laughing or coughing it recognised. That is the whole record of the night. The sound recognition runs on your phone and the raw audio stays there.
We compare each minute with the rest of the same night, because rooms and phones differ wildly. A quiet minute is quiet for you. We smooth the movement over a few minutes either side, mark each minute awake, restless or restful, and then look for the first long stretch of sleep and the last one. Before the first and after the last you are awake in bed, whatever the sensor says. A wake-up has to last at least 3 minutes to count.
The same code runs on your phone for the instant morning report and on our server when the night is uploaded, so the two always agree and a night cannot be posted that the data does not support.
Everything you see from this is labelled as an estimate, because that is what it is.
Estimated sleep stages
The app draws four levels through the night: awake, REM, light and deep. They are estimates. A phone cannot see brain waves, so it works from three clues that sleep research links to the stages.
- Awake or asleep. Movement sounds, loudness changes and picking the phone up say when you were awake.
- Breathing. Steady, even breathing goes with deep sleep. Irregular breathing with a still body goes with REM.
- The shape of the night. Sleep runs in cycles of about 90 minutes, with deep sleep mostly in the first cycles and more REM towards morning. That shape leads, and breathing nudges it, so you see a few cycles rather than a stage change every few minutes.
It is a fixed set of rules, not a trained model. The same code runs on your phone and on our server.
Example night, estimated
ExampleHow accurate can this be?
Not very, and we say so. The best published audio models, trained on lab sleep studies, reach about 88% for sleep against wake and 59 to 70% for four stages, with deep sleep the weakest (HomeSleepNet, Tran et al., 2023; SoundSleepNet, Hong et al., 2022). Our estimate is simpler than those models, so we expect it to do no better.
Independent studies of phone sleep apps against a lab sleep study have found stage detection unreliable (Bhat et al., 2015; Fino and Mazzetti, 2019). That is why every chart carries the word Estimated, the app never states an accuracy figure, and the stages never count towards your score, your streak or the leaderboard.
If your watch records stages in Apple Health or Health Connect and you let the app read them, it shows the watch's stages instead and names the source. A watch is a better instrument than a phone on a table.
What we don't claim
- Accurate sleep stages. We show estimated stages, always labelled Estimated, and never claim they are accurate. The published ceiling for audio is well short of a lab, and deep sleep is the weakest part.
- Diagnosis. Slothdom cannot tell you that you have sleep apnoea, insomnia or anything else. If the app hears a lot of snoring with gasping, it suggests you talk to a GP, and nothing more.
- Medical accuracy. It is a wellbeing app, not a medical device. A wearable or a lab will be more accurate. We are honest about that, including on the night score.
- That more data means better sleep. Watching your numbers too closely can make sleep worse. The research has a name for it, orthosomnia. So the app rewards turning up and keeping a routine, ranks friends on things you control rather than on deep or REM minutes, and never scolds a bad night.
- Hard targets. Eight hours is a default, not a rule. You set the goal that suits you.
How the night score adds up
The score is out of 100. It leans towards things you can do something about: sleeping enough and keeping a routine. Here is the exact split the app uses.
Enough sleep
Time asleep against your own goal. Nothing at 50 percent of it or less, rising evenly to full marks at the goal, so a short night scores like one.
50
Staying asleep
How much of the time in bed you were asleep, with full marks at around 90 percent and nothing below 60. Each wake-up after the first takes 2.5 points off.
25
Keeping your bedtime
How close you went to bed to your target. Within about 15 minutes is full marks, and it fades over the next hour and a half.
15
Calm night
How much of your sleep was restless. A settled night keeps all of it.
10
What counts as meeting your goal
A night meets your goal, and keeps your streak, when three things are true:
- The night was tracked for at least 3 hours.
- You went to bed within 45 minutes of your target bedtime.
- You slept within 30 minutes of your goal.
You earn a streak freeze every 7 nights in a row and can bank 2. A freeze covers one missed night.
Healthy time and the day plan
Healthy time is the time you are asleep plus the time you are awake and away from entertainment apps. Every minute is one or the other, so it comes to 24 hours minus your time in entertainment apps.
The streak counts days at or above a target. The default is 21 hours, which is the same as a daily budget of 3 hours of entertainment. A lower target, such as 16 hours, would be an eight hour budget that almost nobody would break, so it would not tell friends apart. Your own target can be changed in the app.
A missed day uses a streak freeze if you have one. You earn one for every 7 days in a row, Free banks up to 2 and Pro up to 5.
Why iPhone counts in 15 minute steps
Apple lets an app know when you cross a time limit, but does not hand over your usage minute by minute. So on iPhone the app sets a threshold every 15 minutes on the apps you chose and records each one you cross. A day reads as “at least” that many minutes. Thresholds can fire a little early or late, so the app drops any value that is more than the day could hold. Android reports real minutes.
What this is not
Healthy time is a prompt, not a measurement of how well you live, and it is not a medical figure. We use it for you and your friends and nothing else. Friends see the number and your streak, never which apps you use, and a comparison with everyone only appears once at least 20 people are in the group.
Sources
What we read, and what we took from it. Links go to the paper or the guidance itself.
- Recommended Amount of Sleep for a Healthy Adult: A Joint Consensus Statement of the American Academy of Sleep Medicine and Sleep Research Society
Watson et al., Journal of Clinical Sleep Medicine, 2015
Adults should regularly sleep 7 hours or more a night for good health. Our default goal is 8 hours, and you can change it.
- Sleep regularity is a stronger predictor of mortality risk than sleep duration: A prospective cohort study
Windred et al., Sleep (UK Biobank cohort), 2024
Going to bed and waking at consistent times mattered at least as much as total hours. That is why your bedtime counts towards the streak and the score.
- Management of Chronic Insomnia Disorder in Adults: A Clinical Practice Guideline From the American College of Physicians
Qaseem et al., Annals of Internal Medicine, 2016
Cognitive behavioural therapy for insomnia (CBT-I) is the recommended first treatment. Our restless-night prompts borrow its stimulus-control idea: if you are awake for a long while, get up and come back when sleepy.
- Insomnia: Clinical Knowledge Summary
NICE, National Institute for Health and Care Excellence
UK guidance points to sleep hygiene and CBT-I before medicines, and to a GP for insomnia that lasts. We nudge towards the GP, not towards more tracking.
- Orthosomnia: Are Some Patients Taking the Quantified Self Too Far?
Baron et al., Journal of Clinical Sleep Medicine, 2017
Chasing perfect numbers from a tracker can itself make sleep worse. We show estimates, round them, and use language that rewards routine rather than perfection.
- Is There a Clinical Role For Smartphone Sleep Apps? Comparison of Sleep Cycle Detection by a Smartphone Application to Polysomnography
Bhat et al., Journal of Clinical Sleep Medicine, 2015
A phone app compared with a lab sleep study was not accurate enough to read sleep stages. That is why our stages are labelled Estimated and we never claim accuracy.
- Monitoring healthy and disturbed sleep through smartphone applications: a review of experimental evidence
Fino and Mazzetti, Sleep and Breathing, 2019
Phone apps can estimate when you were asleep and awake reasonably well, but they are weaker at sleep stages and are not diagnostic tools.
- Prediction of Sleep Stages Via Deep Learning Using Smartphone Audio Recordings in Home Environments: Model Development and Validation
Tran et al., Journal of Medical Internet Research (HomeSleepNet), 2023
A model trained on smartphone audio recorded at home, checked against lab sleep studies, is among the best published results. It is the kind of ceiling we point to: about 59 to 70% for four stages, and deep sleep is the hardest to get right.
- End-to-End Sleep Staging Using Nocturnal Sounds from Microphone Chips for Mobile Devices
Hong et al., Nature and Science of Sleep (SoundSleepNet), 2022
Sound alone can separate sleep from wake about 88% of the time in a lab-trained model, and four stages far less well. Our own estimate is simpler than these models, so we expect it to do no better.
- Sleep apnoea
NHS, nhs.uk
Loud snoring with gasping or pauses in breathing is a reason to see a GP. When the app hears a lot of it, that is all it says.