What the MaxDex Energy Model Cannot Do

What does MaxDex not measure?

No heart rate, no heart-rate variability, no blood oxygen, no skin temperature, no respiratory rate. On iPhone the app requests exactly two things from Apple Health: step count and sleep. Everything else is either logged by you or inferred from phone signals.

This rules out a direct comparison with wearables on biometric depth, and we would rather draw that line ourselves than have it drawn for us. If a recommendation depends on autonomic state — recovery scores driven by overnight HRV, for instance — MaxDex is not measuring the thing that produces it.

What follows from the narrow permission scope is a privacy property worth stating: an app that never requests heart rate cannot leak it.

Has the model been clinically validated?

No. There is no validation study against polysomnography, actigraphy or any clinical measure, and no published accuracy figure. The model is built on peer-reviewed frameworks, but our specific implementation of them has not been independently tested.

This is the single most important sentence on the site, so it is worth being unambiguous: "based on published models" and "validated" are different claims, and only the first is true here. Borbély’s two-process model and the Banister systems model are well established. Our constants, weights and the way they combine are engineering choices that seemed reasonable, not findings.

We would rather say this than quote a confident accuracy percentage we have not earned. If that work happens, it will be published here with its methods and its failures intact.

Is anything in the model switched off?

Yes. The cognitive driver, intended to capture screen and cognitive load, currently returns no value and contributes exactly zero to your score. Its weight appears in the model but has no effect today.

Screen-time input is wired out of the app and the algorithms. The scaffolding remains, including a re-centring baseline constant so that enabling it later does not create a sudden discontinuity in everyone’s history.

Relatedly, the engine computes a richer blended "mental energy" curve that the interface does not show. Until screen time ships, mental energy is presented as the sleep score alone, so that the app and the AI never quote two different numbers for the same question.

What can a phone genuinely not detect?

Sleep stages. Phone-only detection produces duration, efficiency, disturbance and awake minutes — not REM, deep or light sleep. It also cannot detect sleep at all if the phone is not with you.

The data model can store staging when a platform provides it, but nothing in the phone-only path produces it. Any product claiming phone-only REM detection is describing something its sensors cannot support.

Detection quality also degrades in specific, predictable ways: a phone left on a desk overnight, a night of heavy phone use in bed, or a device that dies. These produce low-confidence readings, which the model then regresses toward your baseline rather than treating as fact — see how uncertain nights are handled.

Is MaxDex a medical device or an ADHD treatment?

No to both. MaxDex is productivity software. It does not diagnose, treat, cure or prevent any condition, including ADHD and executive dysfunction, and no output should be used as a clinical signal.

Many people who use the app describe ADHD-related task paralysis, and the product is designed with that experience in mind. Designing for an audience is not the same as treating a condition, and we do not conflate the two.

If sleep or energy problems are persistent or severe, that is a conversation for a clinician, and a productivity app is not a substitute for one.

What happens when there is not enough data?

The app says so. A date with no logged data returns a neutral 50 explicitly flagged as having no data, and the interface is gated so a bare circadian default cannot be presented as a real reading or forecast.

This is a deliberate design decision with a visible cost: a new user sees an empty state instead of an impressive-looking chart. An invented number is worse than an honest gap, because it teaches you to trust a reading that means nothing.

The same principle governs the forecast band, which widens from ±3 to ±30 points as it projects further out, and the confidence weighting that pulls uncertain nights toward your baseline instead of letting them swing the score.

Frequently asked questions