MindOS
The science

The honest method.

A short, plain account of how MindOS reads your energy, what it predicts well, and exactly where it does not. We would rather you trust a modest claim than fall for a confident one.

The model, in one line

MindOS fits a small, per-person model to your own history. Each day, your expected energy reverts toward your personal baseline, nudged by your recent inputs:

next energy = your baseline + inertia times (yesterday minus baseline) + sensitivity times today's inputs.

The inputs are your sleep, your morning HRV, your steps, and your screen load, each scored against your own normal. The model leans on a sensible prior early on, then becomes more and more your own as it sees more days.

Four numbers come out of that fit, and they are your fingerprint: your baseline (where your usual sits), your volatility (how wide your days swing), your inertia (how fast you revert), and your sensitivity (how much your behavior moves the next day).

What we checked, and what we found

We re-ran the model on a public longitudinal dataset of real people, and we are reporting the modest result rather than the flattering one.

Next-day point prediction: mean absolute error about 0.60 on a 1 to 5 scale. The naive "predict your own average" baseline: about 0.61. The edge on the precise daily number is real but small. That is the reason we never show a daily score.

The value is not in the decimal. It is in three findings that hold up.

The Rope Theorem

Extreme states pull back toward the middle, like a rope going taut. A day near your floor tends to bounce the next day. A peak rarely repeats back to back. In the validation this reversion call was right roughly two times in three, which makes it the single most reliable forward signal in the model. It is the one thing MindOS will actually predict, and only when you are at an extreme.

The Volatility Law

How predictable you are is itself personal. People whose days swing a lot are harder to call, and people whose days hold steady are easier. So MindOS reads your own volatility and hedges accordingly: a steadier person gets a firmer read, a swingier person gets an honest "this is a loose read." The relationship is moderate, not perfect, and we treat it that way.

The Learning Curve

Accuracy improves as your own data accumulates. Early predictions lean on the prior and are rougher. Over your first weeks the error comes down and the read becomes meaningfully about you. This is why MindOS frames its first weeks as the model coming into focus, not as a finished verdict on day one.

What we will not claim

No accuracy percentage on a daily prediction we cannot back. No certainty about tomorrow. No causal claim that one habit "causes" your energy, only associations in your own days that clear a false-discovery filter and can still be coincidence. No overnight HRV, no clinical sleep staging, no medical claims. When a finding is weak, we say it is weak. When there is nothing to report, that is a real result too.

A description of your data, not a grade of you.

Get the app