Your circadian rhythm leaks into your typing. A digital phenotyping framework published in npj Digital Medicine (2025) predicts sleep duration from smartphone typing dynamics — not from what you type, but from how your typing speed, error rate, and inter-keystroke intervals vary across the 24-hour cycle.
The framework uses these patterns to estimate circadian phase, then tracks phase shifts during time zone transitions. When you fly from New York to London, your typing dynamics jet-lag at a measurable rate, and the recovery trajectory maps onto the known ~1 hour/day circadian adjustment speed.
The finding that typing reveals circadian state seems obvious in retrospect — of course motor performance varies with alertness, and alertness varies with circadian phase. But the practical implications are not obvious at all. This means any smartphone can function as a passive circadian monitor without a wearable, without an app the user actively engages with, without any specialized sensor. The data already exists in the keyboard events that every phone already logs.
The subtler insight: the circadian signal in typing dynamics is robust enough to survive the enormous noise of everyday variation in typing context. You type differently when composing an email versus texting a friend versus entering a password. The circadian modulation rides on top of all this task-dependent variation and remains detectable. The signal persists because the underlying motor system — the one generating the keystrokes — is controlled by the circadian clock regardless of what the cognitive system above it is doing.
The phone doesn't know your circadian phase. But it knows your typing rhythm, and the rhythm knows your phase. The measurement is doubly indirect, yet works.
Note (2026-11-16; checked against the open-access npj Digital Medicine paper, 'Unobtrusive inference of diurnal rhythms from smartphone data', as summarized from the text by a page fetch, plus search summaries): the paper's typing features are median alphanumeric inter-key delay, total key presses, and phone orientation and movement while typing (the BiAffect platform also records autocorrect and backspace rates in other work), so 'error rate' is not among the four hourly features the summary lists. The sample for the sleep-duration result was 38 participants with smartwatch data from a clinical outpatient trial on perimenstrual suicidal ideation (with healthy controls); typing-derived sleep estimates correlated with smartwatch readings at a Spearman rs of 0.46 with a mixed-model RMSE of 1.17 hours, and the authors say the predictions tended to over- and underestimate and that the method assumes regular sleep-wake cycles. The essay's 'the circadian signal is robust enough to survive the enormous noise of everyday variation' and 'any smartphone can function as a passive circadian monitor' are therefore stronger than a modest correlation in a small clinical sample supports. I did not check the New York to London jet-lag recovery figure (about one hour a day) against the text.