Can a ring actually verify creatine? What our thresholds say about sleep efficiency

No. A crossover trial with 14 healthy men tested 20 grams of creatine per day, for 7 days, against sleep efficiency measured by actigraphy. The gap between creatine and placebo was just 0.04 percentage points. Our noise threshold for this variable, built from 650 nights of one body, sits at 0.99% even at its strictest window: a 168-night average. An effect of 0.04 does not clear that bar. There is no honest verdict here. Not for creatine, and not against it.

What does "below the threshold" actually mean?

Our ring does not measure substances. It measures signals. Heart rate, heart rate variability, breathing, movement. From those signals we compute sleep efficiency: the share of time in bed your body actually spends asleep. That number moves on its own, night to night, with nothing in your routine changing at all. Before we judge whether any compound moves the needle, we first measure how much the needle moves by itself. That floor has a name: threshold. An effect sitting below the threshold cannot be told apart from your body's own noise. This does not mean the effect is fake. It means our instrument, at this scale, cannot pull it out of the background.

How we measured it: 650 nights, one body

The threshold is not a theoretical formula. It comes from 650 nights, tracked on the same body, with the same ring, without gaps. From those nights we compute how much sleep efficiency naturally drifts with no intervention at all. That drift builds thresholds at different averaging windows. At 28 nights, the threshold is 2.43 points. At 56 nights, it drops to 1.72. At 84 nights, to 1.4. At 112 nights, to 1.21. At 140 nights, to 1.09. And at 168 nights, the threshold hits its lowest floor: 0.99 percentage points. The logic is simple. More nights in the average means less background noise, and a finer real effect becomes visible. But even at our strictest floor, that 0.99-point threshold is still far wider than the 0.04-point effect published on creatine.

The table: the published result against our threshold

Variable	Creatine (20 g/day, 7 days)	Placebo	Statistic
Sleep efficiency	77.75% ± 5.4	77.71% ± 5.97	t = 0.018, p = 0.98
Total sleep time	363.83 min	357.3 min	p = 0.81
Efficiency difference	0.04 percentage points	—	—
Our threshold at 168 nights	0.99 percentage points	—	—

The trial ran on 14 healthy, active men with a sleep quality score (PSQI) under 5. The design was crossover, double-blind, placebo-controlled, and sleep was tracked by actigraphy. This is not a small study done poorly. It is a well-run study that found a clean null. The 0.04-point gap between groups is, in practice, nothing. Efficiency on creatine came in at 77.75%. Efficiency on placebo came in at 77.71%. Total sleep time barely moved either: 363.83 minutes against 357.3 minutes, with a p-value of 0.81.

Why 0.04 doesn't clear the door

Look at the two central numbers in the table. The published effect is 0.04 percentage points. Our threshold, even at its strictest setting — a 168-night average, the lowest noise floor we measured across 650 nights — sits at 0.99 percentage points. The effect sits far below the threshold. You don't need complex math to see it. 0.04 is a small fraction of 0.99. If we published "creatine improves sleep efficiency," that would be a false positive. If we published "creatine does not improve sleep efficiency," that would also be a false positive, because the trial lacked the power to tell a small real effect apart from no effect at all. Neither claim clears the bar we hold ourselves to. That is why this cell is tagged "does not arbitrate." Not because we skipped the measurement. Because we measured with enough rigor to say, out loud, when the data falls short.

Why publish a null instead of staying quiet?

We could have skipped this cell entirely. Nobody asks for a negative result before it exists. But a null published with its number, before anyone requests it, is different from a null that only shows up as a reply to an objection. The first is identity: this is how we work, watched or not. The second is comfort, and comfort proves nothing. This cell exists because our archive of 18483 measurements, across 31 metrics, over 650 nights, includes what cleared the threshold and what didn't. Publishing only the wins would be its own kind of dishonesty.

The limit

This data does not say creatine does nothing for sleep in general. It says that, for this one variable — sleep efficiency by actigraphy — at this dose and this duration, the published effect is too small for our instrument to separate from the body's own noise. Three things this data does not cover. First, the source trial had 14 participants. That is a small sample, with low statistical power to catch a real effect if one existed but was modest in size. Second, the baseline efficiency of these 14 men was 77.7%, well below the 88% seen in other trials, such as those on magnesium. That starting-point gap matters. You cannot directly compare efficiency percentages across different studies, because each population starts from a different place. Third, this result speaks only to sleep efficiency and total sleep time. It says nothing about strength, cognitive performance, or any other effect creatine may have outside of sleep. Creatine has solid evidence in other areas of the body. This cell neither touches that evidence nor denies it.

What comes next

This file is one piece of a larger catalog. You can browse every protocol we've measured at Mental Work — all protocols, or read the related file: nights-creatine-judge-sleep-latency, where the same creatine dose gets checked against a different sleep variable. If you want the full dataset behind every threshold, unedited, it's open at open science — the dataset: 650 nights, 31 metrics, one body.