Every sleep optimization thread arrives at the same rule eventually: last coffee by 2 p.m. It has the shape of good advice — a number, a mechanism, a clean line you can hold yourself to. I held it for about two years on the strength of nothing but its plausibility, and then I bought a ring that measures things, which is how most of my recent opinions have been destroyed.
The verdict: the 2 p.m. cutoff did almost nothing for how fast I fell asleep and a measurable amount for what happened after — and the variable that moved my numbers most wasn't the clock at all, it was how much caffeine I'd had in total that day.
The myth, stated fairly
The rule comes from a real number. Caffeine's half-life in a healthy adult is somewhere around five hours, with a wide personal spread. Coffee at 2 p.m., bed at 11 p.m.: nine hours, nearly two half-lives, so you're down to roughly a quarter of what you drank. A quarter sounds like nothing. That's the entire argument, and it's why the rule travels so well — it's arithmetic, not vibes, and it survives being shouted over background music in a fifteen-second video.
The part nobody says out loud is what a quarter actually is. If the 2 p.m. cup was a 12-ounce pour-over at around 165 mg, you get into bed with roughly 40 mg still circulating. That's a shot of espresso, at lights out, every night, on purpose. Reframed that way the rule stops sounding like a safety margin and starts sounding like a smaller dose at a worse time.
What I actually did
Twenty-one nights. Three conditions, seven nights each, run in blocks rather than alternating — which was a mistake, and I'll come back to it.
- Same beans throughout: a washed Colombian, 22 g in, 360 g out, one grinder setting, about 165 mg per cup by the roaster's published figure.
- Fixed daily total: two cups, roughly 330 mg, the first always at 7:15 a.m.
- The only deliberate variable: when the second cup landed — noon, 2 p.m., or 4 p.m.
- Bedtime inside a 25-minute window. No alcohol for the full three weeks. No naps, no evening training.
- Measured with an Oura ring, Gen 3, plus a note-card by the bed where I wrote a single number: how awake I felt at lights out, 1 to 5.
What the numbers did
Seven-night averages, second cup at the stated time:
| Metric | Noon cutoff | 2 p.m. cutoff | 4 p.m. cutoff |
|---|---|---|---|
| Sleep onset latency | 14 min | 15 min | 19 min |
| Deep sleep | 71 min | 63 min | 54 min |
| Wake events | 1.8 | 2.3 | 3.1 |
| Resting heart rate | 52 bpm | 54 bpm | 56 bpm |
The onset column is the interesting one, because it's the column I would have used to judge the whole thing if I'd had no ring. One minute separates noon from 2 p.m. Five minutes separates noon from 4 p.m. If you asked me at the door of sleep whether caffeine was affecting me, I'd have said no, confidently, on every single night of the 4 p.m. block. My note-card scores don't distinguish the conditions at all.
The rest of the table does. Deep sleep fell 24% from the noon block to the 4 p.m. block, and — this is the part that annoyed me — it fell 11% between noon and 2 p.m., which is the condition I'd spent two years treating as a solved problem.
The thing I got wrong, and the thing I didn't expect
What I got wrong was the outcome I was measuring. I'd been using "I fall asleep fine" as evidence that the cutoff worked. It isn't evidence of anything. Falling asleep is mostly a contest between accumulated sleep pressure and 40 mg of a mild stimulant, and after a normal waking day, sleep pressure wins that contest easily. The caffeine doesn't stop the night from starting. It changes what the night is made of.
The thing I didn't expect showed up in the two weeks after the test, when I stopped controlling timing and started logging dose. My four best deep-sleep nights of the month were all days under 150 mg total, and two of those had coffee after 3 p.m. My four worst were all days over 300 mg, including a noon-cutoff day where I'd front-loaded three cups before 11 a.m. and slept like a person on a train.
If you change one thing, change the size of the second cup, not the hour it lands.
Why this happens
Caffeine works by sitting in adenosine receptors without doing anything. Adenosine builds up across the waking day and is the chemical version of the phrase "I've been up a while"; caffeine blocks the message without clearing the backlog. So the pressure is still there, which is why you still fall asleep — and the blockade is still partly there too, which is why the sleep is shallower than it should be. Slow-wave sleep appears to be more sensitive to residual caffeine than sleep onset is. Those two facts pull in opposite directions, and the one you can feel is the one that doesn't matter.
There's outside support for this. A 2013 trial in the Journal of Clinical Sleep Medicine gave subjects 400 mg at 0, 3, and 6 hours before bed and found meaningful sleep disruption even in the 6-hour condition — measured, not self-reported. The subjects' own ratings underestimated it. Small study, big dose, but the shape matches what my ring kept telling me while my note-card said everything was fine.
What I couldn't test
I am one person with one liver. Caffeine clearance varies enormously between people — the CYP1A2 variation is real and I have no idea which side of it I'm on, so my five-hour half-life assumption may be off by hours in either direction.
I ran the conditions in blocks, not shuffled, so week-over-week drift is baked into the results and I can't separate it out. I knew which condition I was in every night; there was no blinding and no placebo cup. And a consumer ring infers "deep sleep" from heart rate, HRV and movement rather than reading brain activity, so treat that row as directionally useful and nothing more. Twenty-one nights is a gesture at an answer, not an answer.
Who this is for, and who it isn't
Useful to you if: you already track sleep, you've been treating a cutoff time as sufficient, and you've never actually looked at what your total daily dose is. That number is usually larger than people think and nobody posts about it because it isn't a hack.
Not useful to you if: you're a slow metabolizer who already knows an afternoon coffee costs you the night — you don't need my table, you need to stop. Also not useful if you have no measurement at all, because the entire finding here is that the felt experience and the measured one disagree.
Where I landed
My second cup is now 11 grams instead of 22, in a 6-ounce cup I bought specifically because the big one made the small dose look pitiful. It happens at 1:40 p.m., which is not a principled time — it's when I walk back from the mailbox.
