The scores at a glance
Exposure Radar turns everyday signals into four numbers. Each runs from 0 to 100, and all of them use the same colors, from green to red.
Readiness
How ready you are today. Higher is better.
Exposure index
How much you were likely exposed to on one day. Lower is better.
7-day exposure
Your recent exposure, with recent days counting most. Lower is better.
Vitals score
Your vitals compared with your own baseline. 75 is your usual.
This page gives the exact formulas and numbers the app uses. The scores are estimates for personal awareness, not medical measurements, and they can’t tell you whether you’re sick.
Exposure index
Every place you spend time gets a dose: a relative number that grows with the time you spent there and with each factor that makes catching something more likely.
dose = hours × setting × crowding × community spread × climate × mask
A visit that crosses midnight is split between the two days. Each day’s doses are added up and turned into that day’s index, from 0 to 100. The factors are explained below.
A stay with no end. A stay begins when your iPhone reports a place and ends when it reports the next one. If the phone is switched off, left behind or loses Location while you’re somewhere, no end arrives. Away from home or a hotel, the app believes such a stay for 16 hours from when it began, which is longer than a working day or a long shift. After that the time isn’t known, and it adds nothing to the dose: it isn’t scored as more hours at the same place. At home and at a hotel there is no limit, since a phone can sit there for days. The 16 hours are the app’s own choice.
Time nobody logged. Hours that no stay or trip accounts for (before the first place was logged that day, after a stay passed the 16 hours, or while the phone was off) are unknown. They add nothing to that day’s dose, and the day’s row in the list of past days says how many hours were “not logged” when they come to two or more. A day with nothing logged at all has no index: it is left out of the averages below, not counted as a day of zero exposure. A day spent at home is logged, as a stay at home.
Setting: 0.03 to 1.5
The type of place stands in for how well it’s ventilated and how close people usually are. The app looks up the nearest categorized point of interest within 75 meters on Apple Maps. Home is the place where you spend the most nights (a night counts once you’ve been there for two hours between 1 and 5 a.m., so one night is enough to start), or any place you mark as home. The home point is the middle of those nights, and a home area around it still counts as home: 250 meters by default, or 150, 400 or 600 meters, chosen in Settings, so a yard, an apartment complex or a campus is one place rather than several. Stays inside the area are home, and home stays that touch are joined into one. Within 150 meters of the home point a stay is home whatever the map calls it, as before 1.4. A place with a name of its own farther out but inside the area, such as a café down the street, counts as home but stays its own place, so choosing a smaller area gives it back. You can change a place’s type in the app.
| Place type | Setting | Typical people |
|---|---|---|
| Home | 0.03 | 1.25 |
| Outdoors | 0.15 | 1.25 |
| Driving, own car (assumed) | 0.05 | none |
| Walking, running, cycling between places | 0.15 | 1.25 |
| Office | 0.6 | 5 |
| Hotel | 0.6 | 5 |
| Shop | 0.7 | 6.25 |
| Other indoor | 0.7 | 6.25 |
| Food & drink | 1.0 | 9.375 |
| Gym / arena | 1.0 | 9.375 |
| School / library | 1.1 | 9.375 |
| Theater / museum | 1.2 | 12.5 |
| Transit hub | 1.3 | 12.5 |
| Healthcare | 1.5 | 12.5 |
Home is set low because the app can only reflect community exposure there. It can’t see who you live with. “Typical people” is the number of people close by the app assumes when it couldn’t measure one (the device counts of earlier versions, 2 to 20, converted at 1.6 devices per person).
People you live with. With Family & Friends, you can mark a person who shares their scores with you as living with you. Then, for each day you spend at home, a share of the exposure they had in the seven days before counts toward your day: their average daily dose over the days they shared in that week (their daily index, converted back along the same curve) × 0.427 × ½ unless they have a strain signal × the part of a full day you were home (your hours at home that day ÷ 8, at most 1). The 0.427 is the pooled rate at which people living with someone who had Omicron caught it (Madewell et al. 2022); applying it to a dose is the app’s own use of the figure, and the halving is the app’s own choice. The day it comes from is never the same day. A day whose week holds none of their shared days gets no term, so days from before they shared, and weeks they shared nothing for, are untouched; someone who shares only the headline scores and no days has their 7-day load stand in, for the week up to their last update. A strain signal is known only as of their last update, so it counts for that day and a day either side; earlier days count half. Their days are filed by the nearest midnight here, so a sharer in another time zone still has their Tuesday counted as Tuesday. Nothing about them is estimated: the term uses only the daily scores and the strain signal they already share, on your phone. It appears as its own row on the day and under “What drove”, and the same term seen from their side (“What you bring home”, from your own days away) is shown on their page. The scores you share with others are from your own places and don’t include it, so what someone brought home to you never reaches the people you share with, and two people who mark each other don’t feed each other’s scores.
Between the places you visit, the app can use your iPhone’s Motion & Fitness history, if you allow it. The phone’s motion coprocessor classes each stretch of the day as still, walking, running, cycling, in a vehicle or unknown; only the class and the times are read, for the last seven days, and the app keeps on your iPhone each day’s minutes per class and the start and end time of each walk, run, ride or drive of two minutes or more. A stretch of walking, running or cycling that falls in a gap between two stays counts as time outdoors (setting 0.15, light crowding). A stretch in a vehicle counts as a drive in your own car: setting 0.05 and no crowding term, between home and outdoors, because the app can’t see who else is in the car. Buses, trains and shared rides aren’t told apart yet; a stop you add as a Transit hub is scored as one. Motion never changes the scoring of time inside a stay, and time between stays with no movement class (a stop shorter than five minutes, say) stays unscored, as before.
One correction: a logged stay of two hours or less that a drive covers at least 80% of was a roadside stop the location services invented (a light, a drive-through, a slow stretch of road), so it leaves the day’s places and its time counts as part of the drive. Home, a place whose type you set, a place you added by hand and the stay you’re in now are never treated that way; choosing a type for a stop in the app keeps it as a place for good.
Daylight
With an Apple Watch that records Time in Daylight, the part of an indoor stay the watch counted as daylight is weighted like an outdoor place. Each watch sample is spread evenly over its own interval and the overlap with each stay is credited; the share of the stay in daylight blends the place’s weight toward the outdoor weight, 0.15.
setting = (1 − share) × place weight + share × 0.15
An 8-hour office day with a 30-minute lunchtime walk has a share of 30 ÷ 480 = 0.0625, so its setting is 0.6 × 0.9375 + 0.15 × 0.0625 ≈ 0.572 instead of 0.6. Home and outdoor places are not blended (home is already lower, and outdoors is outdoors). Without a watch, or on days the watch has not synced, nothing changes: the stay counts as the kind of place it is. The Indoor public measure on Today is net of the daylight part of indoor stays, and its detail line shows how much was taken off.
Crowding: 0.5 to 2.5
When the app is open during a visit, it listens for 12 seconds to the Bluetooth devices around it and turns them into an estimate of how many people are close by. Each device heard is sorted by what it broadcasts. Apple’s own messages tell an iPhone, Apple Watch or Mac (the app can’t tell these three apart) from AirPods, Find My tags, Apple TVs and HomePods; standard service ids tell heart-rate watches, gym machines, Google Fast Pair earbuds, keyboards, beacons and scales; a Windows PC names its maker; and a device’s own name is read last. Anything that accepts no connection and says nothing recognizable counts as a fixture.
Each kind of device stands for part of a person. An Apple device (an iPhone, Apple Watch, iPad or Mac) counts 0.7, because one person carries about 1.4 of them (a phone, and often a watch or a laptop). A non-Apple phone counts as one person only when it says it is a phone, which today means the retired Exposure Notification service; most Android phones broadcast nothing, and one that does reads as a device the app can’t place. A sports watch or band counts 0.8, since its owner’s phone is often silent. A PC counts 0.5. Earbuds count 0.3, and so does a connectable device the app can’t place, because each is usually beside a phone already counted. A tag counts 0.1 (keys, a bag, a bike, or a sleeping iPhone). TVs, speakers, printers, beacons, lighting, scales, cars and gym machines count 0. These shares are the app’s own estimates.
Nearer devices count more. The middle of each device’s signal readings over the 12 seconds is turned into a rough distance (−59 dBm at 1 meter, falling with the 2.5th power of distance, the usual indoor figure), and a device within 2 meters counts in full, one at 6 meters or beyond not at all, with a straight line between: at 3 meters about three quarters, at 5 meters about a quarter. Signal strength is a noisy guide to distance, which is why the bands are broad. Devices you mark as your own under Settings → My devices are left out.
people = Σ distance weight × person share, over the devices that aren’t yours
crowding = 0.5 + people ÷ 12.5, with people capped at 25
So 0 people gives 0.5, 12.5 gives 1.5, and 25 or more gives 2.5. A visit keeps its highest estimate. If a visit wasn’t scanned, the app uses the typical number of people for that type of place from the table above. At a place you’ve scanned at least twice, an unscanned visit uses the middle of your own estimates there instead. Only the estimate and the counts by kind are kept, never which devices were there. Scans made before version 1.4 counted devices with a strong signal (−70 dBm or better, about 2.75 meters on the same model); those counts are converted at 1.6 devices per person, which gives exactly the scores they had.
Community spread: 0.3 to 2.0
community spread = 0.3 + 1.7 × percentile ÷ 100
The percentile is the combined wastewater level for the county you were in, from 0 to 100 (see Community spread below). A visit keeps the level from the time you were there, so older days aren’t re-scored with today’s levels. With no data, this factor is 1.0.
Climate: 0.7 to 1.4
Dry air helps flu viruses survive and spread (Shaman and Kohn, 2009). The app measures dryness as absolute humidity: grams of water vapor per cubic meter of air (g/m³). It calculates this from the temperature (T, in °C) and relative humidity (RH, in %) that Apple Weather reports for where you were.
absolute humidity = 6.112 × e17.67 T ÷ (T + 243.5) × RH × 2.1674 ÷ (273.15 + T)
outdoors = 1.4 − 0.05 × absolute humidity, kept between 0.7 and 1.4
indoors = 1 + (outdoors − 1) ÷ 2, so between 0.85 and 1.2
Indoors, the effect is halved. Air with 8 g/m³ is neutral (1.0). A cold, dry day (0 °C and 60% humidity, or 2.9 g/m³) gives 1.25 outdoors and 1.13 indoors. A humid summer day (25 °C and 70% humidity, or 16.1 g/m³) gives 0.7 outdoors and 0.85 indoors. Without weather data, this factor is 1.0.
Mask: 0.3
If you note that you wore a mask at a place, that visit’s dose is multiplied by 0.3. Otherwise this factor is 1.0.
From dose to index
daily exposure index = 100 × (1 − e−dose ÷ 10)
The index climbs quickly at first, then flattens as it nears 100. A day’s total dose of 10 gives 63.
| Day’s total dose | Index | Level |
|---|---|---|
| 1 | 10 | Low |
| 3 | 26 | Moderate |
| 5 | 39 | Moderate |
| 10 | 63 | High |
| 15 | 78 | Very high |
| 20 | 86 | Very high |
Levels: under 25 is low, 25–49 moderate, 50–74 high, and 75 or more very high.
A worked example
An evening concert. 3 hours 15 minutes in a concert hall (theater / museum), with about 24 people close by (38 devices, converted at 1.6 per person), local wastewater at the 73rd percentile, and 4.2 g/m³ of absolute humidity.
- hours: 3.25
- setting: 1.2
- crowding: 0.5 + 23.75 ÷ 12.5 = 2.4
- community spread: 0.3 + 1.7 × 0.73 = 1.541
- climate: 1.4 − 0.05 × 4.2 = 1.19 outdoors, so 1 + 0.19 ÷ 2 = 1.095 indoors
- mask: none, so 1.0
dose = 3.25 × 1.2 × 2.4 × 1.541 × 1.095 ≈ 15.8
index = 100 × (1 − e−1.58) ≈ 79, very high
That one visit puts the day at about 79. With a mask, its dose would be about 4.7 and the index about 38.
7-day exposure
Your 7-day exposure is a weighted average of the last seven daily indexes. Each day counts 0.8 times as much as the day after it. A day with nothing logged is left out of both sums: it keeps its place in the week, so the days around it keep their own weights, but it isn’t averaged in as a 0.
7-day exposure = Σ (indexd × 0.8d) ÷ Σ 0.8d, over the logged days, where d runs from 0 (today) to 6 (six days ago)
| Day | Weight | Share of total |
|---|---|---|
| Today | 1.00 | 25% |
| 1 day ago | 0.80 | 20% |
| 2 days ago | 0.64 | 16% |
| 3 days ago | 0.51 | 13% |
| 4 days ago | 0.41 | 10% |
| 5 days ago | 0.33 | 8% |
| 6 days ago | 0.26 | 7% |
The shares in the table are for a week with every day logged. With days missing, the logged days share the whole: if only today and three days ago were logged, today counts 1.00 ÷ 1.51, about 66%, and three days ago about 34%. With no day logged in the week there is no figure, and the screens say so instead of showing 0.
This is the exposure number on the Today screen, on the widget and in Family & Friends.
Community spread
Community spread comes from the National Wastewater Surveillance System (NWSS) run by the U.S. Centers for Disease Control and Prevention (CDC), which follows virus levels in sewage from treatment plants across the country. The app downloads data for COVID-19, influenza A and RSV, plus whether measles was found, for the county you’re in. It finds the county with the FCC’s county lookup, or the U.S. Census Bureau’s geocoder when the FCC is unavailable. If your county has no sewer sites in the data, or its newest samples are more than 21 days old, it uses the sites across your state. Under the wastewater levels, the app shows the date of the newest CDC sample it used.
The app reads the samples that state and local health departments and CDC’s contractor send to CDC. CDC’s data also holds samples from WastewaterSCAN, which are licensed for non-commercial use only (CC BY-NC 4.0); the app leaves those out, so a county that only WastewaterSCAN samples reads from its state. CDC data is available at no charge from CDC. The levels here are worked out by Exposure Radar, aren’t CDC’s, and don’t imply endorsement by the U.S. Government, the Department of Health and Human Services or CDC.
From samples to a percentile
- When a lab doesn’t find the virus, it reports a stand-in value near the smallest amount its test can pick up. The app counts those samples as zero.
- For each sewer site, the app averages the samples from the site’s most recent 14 days. If the virus wasn’t found in most of them, the site is at the 0th percentile (Minimal).
- Otherwise it ranks that average against the levels found at the site over the past year, counting ties as half. A trace found in the off-season therefore ranks low, not high. A site with fewer than eight detections in the past year has too little to rank against, so it reads Minimal.
- A site counts only if it has a sample from the last 21 days and at least 8 samples in all. Sites are weighted by the number of people they serve.
Combining the viruses
combined = 0.6 × highest + 0.4 × average, using the COVID-19, influenza A and RSV percentiles
+ 10 (up to 100) if measles was found in the last 3 weeks
Leaning on the highest value keeps one surging virus from being averaged away.
Levels and trend arrows
| Level | Percentile |
|---|---|
| Minimal | below 20 |
| Low | 20–39 |
| Moderate | 40–59 |
| High | 60–79 |
| Very high | 80 and above |
Measles shows only whether it was found. Trend arrows compare the average of the latest 14 days with the 14 days before: up if it’s more than 15% higher, down if it’s more than 13% lower.
Vitals score
The vitals score compares seven Apple Health signals with your own recent history. Overnight signals (respiratory rate and wrist temperature) and sleep count toward the morning you wake up. The score needs at least one core signal (resting heart rate, heart rate variability, respiratory rate or sleeping wrist temperature); steps, sleep and blood oxygen refine it but can’t make it on their own.
Which reading counts as now. A day is scored only if it has a core signal, so a day without your wearable has no score. The score shown as current on Today, the widget, the watch and in Family & Friends, and the one readiness and event plans use, is today’s, or yesterday’s when today has none yet (before the night’s data has synced, yesterday’s is the newest there can be). If the newest scored day is older than that, there is no current score: the app says “No recent vitals”, the Vitals tab gives the date of the last reading, readiness is worked out as if the vitals score were at your baseline (75), as it is before any vitals exist, and an earlier strain signal is no longer shown or shared as if it were today’s. Past days keep their scores in the chart.
| Signal | Strain when | Weight |
|---|---|---|
| Resting heart rateminimum spread 1.5 bpm | higher | 1.0 |
| Heart rate variability (SDNN)minimum spread 4 ms | lower | 1.0 |
| Sleeping wrist temperatureminimum spread 0.15 °C | higher | 1.0 |
| Respiratory rate while asleepminimum spread 0.4 breaths/min | higher | 0.9 |
| Blood oxygenminimum spread 0.7 percentage points | lower | 0.6 |
| Sleepminimum spread 0.4 hours | lower | 0.6 |
| Stepsminimum spread 800 steps | lower | 0.3 |
Step 1: your baseline
For each signal, the baseline is the median of your daily values from 60 days to 4 days before the day being scored. The 3 most recent days are left out, so a change that’s just starting doesn’t become part of its own baseline. At least 10 days of data are needed, which in practice means about two weeks of wearing Apple Watch.
Step 2: how far from your usual
z = (today’s value − baseline median) ÷ spread
spread = the larger of 1.4826 × MAD and the signal’s minimum spread
MAD is the median absolute deviation: the typical distance of your daily values from their median. Unlike a standard deviation, one odd day can’t throw it off. Multiplying it by 1.4826 puts it on the same scale as a standard deviation. The minimum spread keeps a very steady baseline from turning tiny changes into big numbers.
Step 3: point every signal the same way
Each z-score is flipped where needed, so that a positive number always means a change in the direction linked to strain: higher resting heart rate, respiratory rate or temperature, or lower heart rate variability, blood oxygen, sleep or steps. Each adjusted value is then kept between −2 and +4. That way one extreme reading can’t dominate, and better-than-usual readings can lift the score only a little.
Step 4: the score
vitals score = 75 − 15 × (weighted average of the adjusted values), kept between 0 and 100
Only signals with data that day are included, and the weights are shared out among them. A score of 75 means you’re at your baseline. Because of the −2 to +4 limits, the lowest possible score is 15.
- 80–100: above baseline
- 65–79: at baseline
- 50–64: mild strain
- under 50: elevated strain
Example. Your resting heart rate this morning is 64 bpm. Your baseline median is 58 bpm, with a MAD of 1.5 bpm.
spread = the larger of 1.4826 × 1.5 = 2.22 and 1.5, so 2.22
z = (64 − 58) ÷ 2.22 ≈ 2.7
Say heart rate variability is also down, with an adjusted value of 2.0, and the other five signals are at baseline (0). All seven weights add up to 5.4.
weighted average = (1.0 × 2.7 + 1.0 × 2.0) ÷ 5.4 ≈ 0.87
vitals score = 75 − 15 × 0.87 ≈ 62, mild strain
Strain signal
A core signal is “off” when its adjusted value is at least 1.5 (at least 1.0 for sleeping wrist temperature). The four core signals are resting heart rate, heart rate variability, respiratory rate and sleeping wrist temperature.
The app shows a strain signal, “Your vitals show signs of strain”, when either:
- two or more core signals are off today, and two or more were off yesterday; or
- three or more core signals are off today.
When new Health data arrives, the app checks in the background and sends a notification, at most once every three days. In the example above, resting heart rate (2.7) and heart rate variability (2.0) are both off. If two core signals were also off the day before, that’s a strain signal.
These vitals shift with alcohol, vaccines, hard training, a short night, stress, the menstrual cycle and feeling unwell (see Sources). A strain signal means your vitals have changed compared with your usual, not that anything is wrong. The app doesn’t diagnose any disease, and it can’t tell you whether you have one or will get one.
Two things to keep in mind. You may get a strain signal when you feel fine, or feel unwell without one; if you’re unsure, trust how you feel. And outside a local surge, most strain signals are not something you caught: these vitals move with everyday life far more often than with anything catching. Because the signal compares you with your own recent past, it isn’t meant for pregnancy, or for a medical condition or a medicine that changes these vitals.
Readiness
readiness = 0.6 × vitals score + 0.4 × (100 − 7-day exposure)
Your vitals describe how your body is doing now, so they count for more. Exposure is only what you were around, and it may never come to anything.
- During a strain signal, readiness is held at 40 or below.
- Until your vitals baseline exists, the app assumes typical vitals: a vitals score of 75.
- Readiness appears once at least one place has been logged.
- The app also tells you what’s holding readiness back most: a vitals score below 75, or 7-day exposure above 25, whichever costs more points.
Levels: 75–100 is Ready, 60–74 Mostly ready, 45–59 Take it easy, and under 45 Rest and recover.
Example. The vitals score is 78 and 7-day exposure is 52.
readiness = 0.6 × 78 + 0.4 × (100 − 52) = 46.8 + 19.2 = 66, mostly ready
Event planning
When you choose an event from your calendar, the app sets a low-exposure window before it: 5 days by default, adjustable from 2 to 21. Five days covers the usual incubation time: a median of about 1.4 days for influenza A and 4.4 days for RSV (Lessler 2009), and a mean of about 3.4 days for COVID-19 with Omicron (Wu 2022).
- Target: keep each day’s exposure index under 30 during the window.
- Window average and your typical day: the average shown for the window, and the typical day it is compared with (the median of your last 30 logged days’ indexes), use logged days only. If days of the window have nothing logged, the plan says how many, because a missing day isn’t a day under the target.
- Reminders: at 9 a.m. on the day the window starts, and 14 days before the event, because flu and COVID shots take about two weeks to reach full protection.
- Destination: if the event has a location, the app finds its county and shows the wastewater levels there.
Caution level
Under Settings → Caution level you can ask the app to be more careful for you: Usual, More careful or Much more careful, and separately say that you take medicine that changes your heart rate or temperature. The screen says who tends to pick each level (people in a higher-risk group, or caring for someone whose immunity is lowered, tend to pick More careful; people whose immunity is lowered tend to pick Much more careful), but the app keeps the level you chose, never the reason. It stays on your phone and the watch paired with it: the setting is never shared, never sent to me and never part of the analytics, though the vitals score you share is worked out with it.
The exposure index doesn’t change with it. It describes what you were around, and two people in the same room on the same afternoon get the same number. What changes is the app’s own lines: the target before an event is under 25 instead of 30 at More careful and under 20 at Much more careful, which also starts new events with a 7-day window. In dose terms a target of 20 treats the same day as counting about 1.6 times as much, and 25 about 1.24 times. Those are the app’s choices, not measured quantities: the evidence that lowered immunity raises the chance of catching something is modest (at most about 1.5 times, and confounded by testing), while the evidence that it raises the severity of an illness is strong, and the lower targets are a margin for that. An ordinary office day at median community spread scores about 37 by the model’s own calibration, over every target, so at Much more careful the target is in practice a mask target: a mask cuts a stay’s dose to 0.3 of what it would be, which brings that day to about 13. The strain signal keeps the same thresholds and two-day rule for everyone. At Much more careful, two core signals off on one night are shown as “worth a check” in the app, never as a notification. Medicine that holds heart rate or temperature down, lowered immunity and older age can keep the signals quieter than usual, so at Much more careful a normal sleeping temperature, and with the medicine switch a normal resting heart rate or temperature, no longer counts in the score’s favour, and the app says beside a calm reading that it’s weak reassurance. Because the vitals score is a weighted average of those signals, that can read a few points lower than it otherwise would; the vitals score is already a comparison with your own baseline, so it was never comparable between people the way the exposure index is. At the careful levels a strain signal adds one line: that you may want to check in with your doctor sooner than you otherwise would. It names no illness and no treatment and sets no deadline, because the signal isn’t evidence enough for an instruction.
Limitations
- Estimates, not medical measurements. The app can’t tell you whether you’ve caught something or could pass it on.
- Wastewater is community-level. It describes a whole area, not you or the people near you. It lags by days, and not every county has sewer sites.
- U.S. data. Community spread uses U.S. data sources. Where there’s no data, that factor is neutral (1.0).
- Bluetooth estimates are a proxy. Telling what a device is from what it broadcasts is a set of rules of thumb; one person may carry several devices, or none; phones that don’t broadcast, which many Android phones don’t, are missed, so the number runs low in some crowds; and scans happen only while the app is open.
- Place type is inferred from nearby map points, so it can be wrong. You can correct it.
- Location can miss visits, especially without “Always” and Precise Location. Time the app has no record of is treated as unknown, not as time at home, so a day with gaps can read lower than it was.
- Daylight needs a capable watch. Time in Daylight comes from Apple Watch’s light sensor, so a sunny window seat or a car can count as daylight and an evening outdoors after dark does not. Days before the watch syncs count as indoor.
- Driving is assumed to be a private car. The app can’t see who else is in it, and buses and trains aren’t told apart. Motion history covers only the last seven days when the app reads it, so a week in which the app isn’t opened or refreshed in the background leaves a gap, and the phone’s classes can lag a few minutes behind.
- Vitals move for many reasons, including exercise, alcohol, travel and stress, and they depend on how consistently you keep your wearable on.
- A simplified model. The factor values are the developer’s estimates, informed by the sources below. The app hasn’t been clinically validated.
Model changes
- 1.5: a stay with no end is believed for 16 hours away from home or a hotel, where it used to run until the next place was reported, so a phone that died somewhere no longer scores that place for days. Days with nothing logged are left out of 7-day exposure, the event window’s average and the typical day, where they used to count as 0. For someone with unlogged days in the past week, 7-day exposure reads higher than it did, and readiness lower; days and weeks that were fully logged score as before. The vitals score shown as current is now today’s or yesterday’s only; before, the last scored day in the past 30 was shown, and shared, as today’s.
- Caution level (version 1.5): an optional setting that lowers the target before events, lengthens the default window at Much more careful, shows a one-night “worth a check”, and stops a normal reading on a blunted signal from raising the vitals score. Nothing changes for anyone who leaves it at Usual, and the exposure index never changes.
- People you live with (version 1.5): with Family & Friends, a share of the exposure someone you live with had in the week before counts toward your days at home, from the daily scores they share. Nothing changes until you mark someone as living with you, and the scores you share with others don’t include it.
- 1.4: home area, daylight blend, people-based crowding, driving. On the first launch after the update the app sorts your saved places once for the home area: stays within 250 meters of home now count as home and pieces of one stay at home are joined, so past days with time near home can score lower (rarely higher, when a recent stay the app had called home lies just outside the area); changing the home area in Settings sorts them again. With an Apple Watch, the last two weeks of stays get their time in daylight, which can only lower them. When you set up Motion & Fitness, the app also reads the week before, so drives and walks between places on those days count too. Otherwise days before the update keep their old scores; week comparisons pause for two weeks after Motion & Fitness is set up.
Sources
The published research and public data behind each score. The app lists the same sources under Settings → How it’s measured → Sources, and on each score’s screen.
Exposure index
- Riley EC, Murphy G, Riley RL (1978). “Airborne spread of measles in a suburban elementary school.” American Journal of Epidemiology 107(5):421–432. doi:10.1093/oxfordjournals.aje.a112560
Used for: turning a day's dose into the 0–100 index - Bulfone TC et al. (2021). “Outdoor transmission of SARS-CoV-2 and other respiratory viruses: a systematic review.” Journal of Infectious Diseases 223(4):550–561. doi:10.1093/infdis/jiaa742
Used for: outdoors counting far less than indoors, and the daylight part of an indoor stay counting like outdoors: the review found the odds of indoor transmission 18.7 times those outdoors (95% CI 6.0–57.9), with under 10% of reported cases caught outdoors - Chang S et al. (2021). “Mobility network models of COVID-19 explain inequities and inform reopening.” Nature 589:82–87. doi:10.1038/s41586-020-2923-3
Used for: the kind of place and how crowded it is - Leclerc QJ et al. (2020). “What settings have been linked to SARS-CoV-2 transmission clusters?” Wellcome Open Research 5:83. doi:10.12688/wellcomeopenres.15889.2
Used for: which kinds of places count more - World Health Organization (2021). “Roadmap to improve and ensure good indoor ventilation in the context of COVID-19.” WHO. www.who.int/publications/i/item/9789240021280
Used for: ventilation, the reason indoor places count more - Leith DJ, Farrell S (2020). “Coronavirus contact tracing: evaluating the potential of using Bluetooth received signal strength for proximity detection.” ACM SIGCOMM Computer Communication Review 50(4):66–74. doi:10.1145/3431832.3431840
Used for: Bluetooth signal strength as a rough, noisy stand-in for distance: why nearer devices count more only in broad bands - Andrejko KL et al. (2022). “Effectiveness of face mask or respirator use in indoor public settings for prevention of SARS-CoV-2 infection.” MMWR 71(6):212–216. doi:10.15585/mmwr.mm7106e1
Used for: the mask factor: about 0.3, close to the lower odds of catching the virus reported with surgical masks - Shaman J, Kohn M (2009). “Absolute humidity modulates influenza survival, transmission, and seasonality.” PNAS 106(9):3243–3248. doi:10.1073/pnas.0806852106
Used for: the air dryness factor: low absolute humidity helps flu survive and spread - Shaman J et al. (2010). “Absolute humidity and the seasonal onset of influenza in the continental United States.” PLoS Biology 8(2):e1000316. doi:10.1371/journal.pbio.1000316
Used for: the air dryness factor: flu seasons start after spells of low absolute humidity - Bolton D (1980). “The computation of equivalent potential temperature.” Monthly Weather Review 108(7):1046–1053. doi:10.1175/1520-0493(1980)108<1046:TCOEPT>2.0.CO;2
Used for: absolute humidity from temperature and relative humidity - Qian H et al. (2021). “Indoor transmission of SARS-CoV-2.” Indoor Air 31(3):639–645. doi:10.1111/ina.12766
Used for: daylight time counting like outdoors: all 318 outbreaks of three or more cases in one review of Chinese case reports happened indoors - Apple (2025). “See time spent in daylight on Apple Watch.” Apple Watch User Guide. support.apple.com/guide/watch/apd3ab22534c
Used for: time in daylight: the Apple Watch light sensor's estimate of time outdoors, read from Apple Health - Longo E, Redondi AEC, Cesana M (2019). “Accurate occupancy estimation with WiFi and Bluetooth/BLE packet capture.” Computer Networks 163:106876. doi:10.1016/j.comnet.2019.106876
Used for: counting the Bluetooth devices people carry to estimate how many people are in a space - Martin J et al. (2019). “Handoff All Your Privacy: A Review of Apple’s Bluetooth Low Energy Continuity Protocol.” Proceedings on Privacy Enhancing Technologies 2019(4):34–53. doi:10.2478/popets-2019-0057
Used for: what iPhones, Apple Watches and AirPods broadcast, which lets the app tell them from fixtures - Weppner J, Lukowicz P (2013). “Bluetooth based collaborative crowd density estimation with mobile phones.” IEEE PerCom 2013:193–200. doi:10.1109/PerCom.2013.6526732
Used for: crowd density from Bluetooth device counts taken by phones - Hu M et al. (2021). “Risk of Coronavirus Disease 2019 Transmission in Train Passengers: an Epidemiological and Modeling Study.” Clinical Infectious Diseases 72(4):604–610. doi:10.1093/cid/ciaa1057
Used for: public transport scored as a transit setting: on trains, risk rose with every hour spent near a passenger who was carrying the virus - Shen Y et al. (2020). “Community Outbreak Investigation of SARS-CoV-2 Transmission Among Bus Riders in Eastern China.” JAMA Internal Medicine 180(12):1665–1671. doi:10.1001/jamainternmed.2020.5225
Used for: why a bus isn't scored like your own car: a third of the riders on a bus with recirculated air caught the virus, wherever they sat - Mathai V, Das A, Bailey JA, Breuer K (2021). “Airflows inside passenger cars and implications for airborne disease transmission.” Science Advances 7(1):eabe0166. doi:10.1126/sciadv.abe0166
Used for: the private-car assumption: a car counts low only because the app takes you to be alone or with your household; with a passenger, open windows change the risk - Madewell ZJ, Yang Y, Longini IM, Halloran ME, Dean NE (2022). “Household Secondary Attack Rates of SARS-CoV-2 by Variant and Vaccination Status: An Updated Systematic Review and Meta-analysis.” JAMA Network Open 5(4):e229317. doi:10.1001/jamanetworkopen.2022.9317
Used for: the household share, 0.427: the pooled rate at which people living with someone who had Omicron caught it (42.7%; Alpha 36.4%, Delta 29.7%). The app applies it to the daily dose a person shares with you, which is its own use of the figure - Apple (2026). “CMMotionActivityManager.” Apple Developer Documentation. developer.apple.com/documentation/coremotion/cmmotionactivitymanager
Used for: where the walking, cycling and driving classes come from: the phone's motion coprocessor, which keeps seven days and reports with a delay of a few minutes
Community spread
- U.S. Centers for Disease Control and Prevention (2026). “Wastewater Monitoring Program (National Wastewater Surveillance System).” CDC. www.cdc.gov/wastewater/
Used for: where the wastewater levels come from - U.S. Centers for Disease Control and Prevention (2026). “CDC Wastewater Data for SARS-CoV-2.” data.cdc.gov, dataset j9g8-acpt. data.cdc.gov/d/j9g8-acpt
Used for: COVID-19 levels - U.S. Centers for Disease Control and Prevention (2026). “CDC Wastewater Data for Influenza A.” data.cdc.gov, dataset ymmh-divb. data.cdc.gov/d/ymmh-divb
Used for: flu A levels - U.S. Centers for Disease Control and Prevention (2026). “CDC Wastewater Data for RSV.” data.cdc.gov, dataset 45cq-cw4i. data.cdc.gov/d/45cq-cw4i
Used for: RSV levels - U.S. Centers for Disease Control and Prevention (2026). “CDC Wastewater Data for Measles.” data.cdc.gov, dataset akvg-8vrb. data.cdc.gov/d/akvg-8vrb
Used for: whether measles was found - U.S. Centers for Disease Control and Prevention (2026). “NSSP Emergency Department Visit Trajectories by State and Sub State Regions: COVID-19, Flu, RSV, Combined.” data.cdc.gov, dataset rdmq-nq56 (National Syndromic Surveillance Program). data.cdc.gov/d/rdmq-nq56
Used for: emergency visits for COVID-19, flu and RSV
Vitals score
- Leys C et al. (2013). “Detecting outliers: do not use standard deviation around the mean, use absolute deviation around the median.” Journal of Experimental Social Psychology 49(4):764–766. doi:10.1016/j.jesp.2013.03.013
Used for: your baseline: the median and median absolute deviation - Plews DJ et al. (2013). “Training adaptation and heart rate variability in elite endurance athletes: opening the door to effective monitoring.” Sports Medicine 43(9):773–781. doi:10.1007/s40279-013-0071-8
Used for: comparing HRV with your own rolling baseline, as studied in athletes - Shaffer F, Ginsberg JP (2017). “An overview of heart rate variability metrics and norms.” Frontiers in Public Health 5:258. doi:10.3389/fpubh.2017.00258
Used for: heart rate variability (SDNN) - Prather AA et al. (2015). “Behaviorally assessed sleep and susceptibility to the common cold.” Sleep 38(9):1353–1359. doi:10.5665/sleep.4968
Used for: sleep counting toward the score - Jung H et al. (2023). “Validating a consumer smartwatch for nocturnal respiratory rate measurements in sleep monitoring.” Sensors 23(18):7976. doi:10.3390/s23187976
Used for: breathing rate during sleep from a watch - Apple (2025). “Track your overnight and daytime vitals with Apple Watch.” Apple Support. support.apple.com/120142
Used for: the overnight vitals Apple Watch records, which aren't intended for medical use
Strain signal
- Pietilä J et al. (2018). “Acute effect of alcohol intake on cardiovascular autonomic regulation during the first hours of sleep in a large real-world sample of Finnish employees.” JMIR Mental Health 5(1):e23. doi:10.2196/mental.9519
Used for: alcohol raising overnight heart rate and lowering HRV - Quer G et al. (2022). “Inter-individual variation in objective measure of reactogenicity following COVID-19 vaccination via smartwatches and fitness bands.” npj Digital Medicine 5:49. doi:10.1038/s41746-022-00591-z
Used for: resting heart rate rising for a few days after a vaccine, peaking on day 2 - Maijala A et al. (2019). “Nocturnal finger skin temperature in menstrual cycle tracking: ambulatory pilot study using a wearable Oura ring.” BMC Women's Health 19:150. doi:10.1186/s12905-019-0844-9
Used for: sleeping temperature changing across the menstrual cycle - Alavi A et al. (2022). “Real-time alerting system for COVID-19 and other stress events using wearable data.” Nature Medicine 28:175–184. doi:10.1038/s41591-021-01593-2
Used for: signals from several vitals at once, and how many follow stress, sleep or alcohol - Radin JM et al. (2021). “Assessment of prolonged physiological and behavioral changes associated with COVID-19 infection.” JAMA Network Open 4(7):e2115959. doi:10.1001/jamanetworkopen.2021.15959
Used for: resting heart rate taking weeks to settle after symptoms start
Event planning
- Lessler J et al. (2009). “Incubation periods of acute respiratory viral infections: a systematic review.” Lancet Infectious Diseases 9(5):291–300. doi:10.1016/S1473-3099(09)70069-6
Used for: the 5-day window: median incubation of about 1.4 days for flu A and 4.4 for RSV - Wu Y et al. (2022). “Incubation period of COVID-19 caused by unique SARS-CoV-2 strains: a systematic review and meta-analysis.” JAMA Network Open 5(8):e2228008. doi:10.1001/jamanetworkopen.2022.28008
Used for: the 5-day window: mean incubation of about 3.4 days for COVID-19 (Omicron) - U.S. Centers for Disease Control and Prevention (2025). “Key facts about seasonal flu vaccine.” CDC. www.cdc.gov/flu/vaccines/keyfacts.html
Used for: antibodies developing about two weeks after a flu shot
Exposure Radar isn’t affiliated with or endorsed by these authors or organizations.