Your two o'clock coffee is still awake at midnight.
Caffeine has a half-life of about five hours, which means half of it is still in you five hours later, and a quarter of it ten hours later. Most people who sleep badly and have "cut back on caffeine" are still drinking it far too late. Here is the arithmetic, drawn.
0mg
still in you at lights out
That is a full coffee's worth still circulating at lights out. Expect a longer time to fall asleep and less deep sleep in the first half of the night, whether or not you feel wired.
Roughly a cup of tea's worth is still on board. Enough to lighten early sleep for most people. Moving the last dose ninety minutes earlier usually clears it.
A small amount is left, unlikely to matter for most people. If you are sensitive or you fall asleep slowly, pull the last dose an hour earlier and see whether anything changes.
Effectively cleared by bedtime. Caffeine is not what is standing between you and sleep tonight.
The half-life is the whole story. Caffeine does not stop working when you stop feeling it. It keeps occupying adenosine receptors, which is why a coffee you barely noticed at two in the afternoon can still cost you deep sleep.
Sleep loss from caffeine is mostly invisible from the inside. In controlled work, people given caffeine six hours before bed lost over an hour of sleep and did not reliably notice.
Real clearance ranges from about 1.5 to 9.5 hours depending on genetics, hormonal contraceptives, pregnancy, smoking and liver function. Treat the band as the answer, not the milligram. Nothing here is saved or sent anywhere.
Scientific Receipt, Caffeine follows first-order elimination with a mean half-life near five hours in healthy adults, though the range across individuals is roughly 1.5 to 9.5 hours depending on CYP1A2 activity, oral contraceptive use, pregnancy and smoking (Institute of Medicine, Caffeine for the Sustainment of Mental Task Performance, 2001). In a controlled trial, 400mg taken six hours before bed reduced total sleep time by more than an hour, and participants did not reliably detect the disruption themselves (Drake et al., Journal of Clinical Sleep Medicine, 2013). Read the full evidence review →