Can an app predict burnout?

We build one of these apps, so treat this page with the appropriate suspicion. That is exactly why it is worth writing down the limits explicitly — a category full of confident wellbeing scores with no stated method deserves a page that says what the ceiling is.

What a phone can actually measure

Modern phones and watches, via Apple Health, give reasonable access to a handful of things that research links to chronic stress:

What a phone cannot measure — at all

• Cortisol, inflammation, or anything requiring a blood sample. No wrist sensor reads these.
• Your job. Autonomy, support and psychological safety are among the strongest predictors of burnout symptoms, and a phone knows nothing about them unless you type them in.
• Depression. That is a clinical assessment. See burnout vs depression.
• Burnout itself. Because there is no criterion standard to measure against, no consumer app has been validated against a diagnosis. Nobody's has.

The baseline problem — the thing most apps get wrong

Between-person variation is enormous compared with the effect being detected. HRV can differ threefold between two equally healthy adults; habitual sleep need varies; resting heart rate spans a wide range. Score someone against a population average and you will tell half your users they are in crisis and the other half that they are fine, for reasons that have nothing to do with stress.

The only defensible approach is within-person: build each user's own rolling baseline and measure the deviation from it. Statistically this argues for robust estimators — a median and a median absolute deviation rather than a mean and a standard deviation — so that one wild night after a flight does not shift the reference for a fortnight. And each signal's influence should be capped, so no single bad reading can swing the whole score.

What an honest forecast looks like

A 7-day projection from two weeks of noisy daily data carries real uncertainty. Presenting it as a confident line is a design decision made against the user. An honest forecast:

• Shows an uncertainty band that widens as the horizon extends.
• Says how confident it is, and says so loudly when the answer is "not very".
• Refuses to pretend during the first week, when there is no baseline yet.
• Is framed as a trend continuing, not destiny. It is a straight line through your recent days, and any change you make invalidates it — which is the point.

Seven questions to ask before trusting any wellbeing score

1. Is the method published? If you cannot find out how the number is computed, it is a mood ring.
2. Does it compare me to myself? Population percentiles are close to meaningless for HRV.
3. Does it show what is driving the score? A number with no attribution cannot be acted on.
4. Does it state its uncertainty? Confidence without calibration is a red flag.
5. Does it cite sources for health claims? Or does it assert that a number means something with nothing behind it?
6. Where does my data live? On-device with no account is a strong answer. An account requirement means the data leaves.
7. Does it claim to diagnose? If it does, it is wrong, and everything else it says is suspect too.

How Pulse answers those seven

In the interest of being judged by our own checklist:

• Method: published in the app, under Settings → Method & sources. Robust z-scores against your 30-day median, direction-normalised, capped at ±3, weighted, mapped to 0–100, smoothed across days.
• Baseline: yours. A first calibration week is marked as low confidence rather than guessed at.
• Attribution: the top contributors are listed and ranked every day.
• Uncertainty: the 7-day forecast is a regression over your last 14 days, drawn with a 95% band that widens with the horizon.
• Sources: every health claim in the app links to its published source. The weights themselves are our editorial judgement informed by that literature — not a validated clinical instrument, and the app says so.
• Data: on-device SQLite. No account, no cloud sync, no analytics.
• Claims: Pulse is an early-warning tool for wellbeing, not a medical device. It does not diagnose, treat or prevent anything.

When to close the app

Tracking is for drift. If a score is making you anxious, if you are checking it several times a day, or if a bad number ruins a day that was otherwise fine, the tool has become the problem — stop using it. And if your symptoms are severe, persistent, or extend beyond work, that is a conversation with a doctor, not with a chart.