Spontaneous Waking Precedes the Alarm by a Fixed Lead Under Randomised, Concealed Schedules: 47,318 Nights of Polysomnography, a Sham-Setting Trial, and a Lead-Compensator Model of Sleep Timing

You wake up a minute before your alarm. We show that this is not habit but a lock: sleep timing tracks the time entered into the alarm device, whether or not the sleeper is ever told what that time is. Across 47,318 nights on which a sealed bedside module drew each night’s alarm time at random and never displayed it, waking preceded the alarm by $61.4\mathrm{s}$ (95% CI $58.9$–$63.9\mathrm{s}$). There was no secondary peak at any customary hour, and no comparable structure on nights when no alarm was set. The lead did not move with the hour set, with accumulated sleep debt, with age, or with the previous evening’s alcohol; a lead that is constant across every hour tested is a phase advance, not a remembered interval. We then disconnected the sounder. Participants who performed the identical setting ritual on a device that could not ring woke, in silence and in darkness, roughly a minute before a time that was drawn, recorded, and never announced; silent nights were indistinguishable from ringing ones. The alarm is therefore not the cause of the waking but an event that follows it, and the operative act is the registration. Sleep does not keep time. It keeps appointments.