Can a ring actually verify ashwagandha (Withania somnifera)? What our thresholds say about sleep latency

The most-cited ashwagandha sleep trial found a 2-minute drop in latency. Active group: 10 minutes. Placebo: 12 minutes. That is -16.7%, measured by actigraphy in 150 healthy adults. Our threshold at 168 nights is 18.53%. The published effect does not reach it. No verdict.

The finding: 2 minutes across 150 people

Deshpande 2020, published in Sleep Medicine, randomized 150 healthy adults to 120 mg/day of Shoden extract (42 mg withanolides) or placebo for 42 days. The primary outcome: sleep onset latency by actigraphy.

At week 6, the active group fell asleep in 10 minutes. The placebo group took 12 minutes. The gap: 2 minutes. In relative terms: -16.7%.

Actigraphy, in this context, records body movement across nights and estimates the moment motor activity ceases. It is a measure of observable behavior, not of the internal physiology of sleep architecture.

That is where the problem starts.

The gap: a group average is not your night

A -16.7% result across 150 people means the active group had shorter latencies on average. It does not say what happened in each individual. It does not say whether half responded strongly and the other half saw nothing. There is no record to repeat the result in a single body.

That is hole H3 in our framework: real group effect, not individualized.

To individualize it, you would need the response distribution for each participant: who improved, who worsened, who saw no change. That level of detail is not available in the published report.

The trial also reported subjective improvement: 72% of the active group said sleep quality improved. The placebo reported 29%. That gap is wide. The actigraphy recorded 2 minutes. The distance between what people felt and what the device measured is large.

When the subjective-objective gap is this wide, the effect can be real and the instrument can be too noisy to capture it in one person. Both can be true at once.

What your morning data faces

You take ashwagandha for 42 days. You check your ring. Did latency drop?

The number you see carries its own noise. Not the trial's noise — the noise of one body. Measured across 650 nights of case zero. Our sleep latency has a coefficient of variation of 84.41%. The relative coefficient of variation of change is 233.96%. A change must be large to emerge from that noise.

The trial showed -16.7%. Our threshold at 168 nights is 18.53%. The published effect falls below the threshold — even at the longest follow-up we model.

There is no horizon within the current model where that percentage crosses the line.

The threshold table

From ruido.medido.json. Measured across 650 nights, single subject. 18483 total records.

Follow-up nights	Threshold for an honest verdict
28	45.38%
56	32.09%
84	26.2%
112	22.69%
140	20.29%
168	18.53%

Ashwagandha's published effect on sleep latency: -16.7%. It does not reach any threshold at any horizon.

What "does not arbitrate" means

"Does not arbitrate" is not a negative verdict. It means the instrument cannot give an honest verdict for that effect size.

Three situations exist when an effect falls short of a threshold. First: the compound has no real effect in this body. Second: the effect is real, but the variable's noise hides it. Third: the effect exists but needs more follow-up than the model covers.

Sleep latency has a coefficient of variation of 84.41%. It is our noisiest variable. Even real shifts of 6 or 8 minutes can be lost inside night-to-night noise.

The series autocorrelation is 0.026. Last night's latency barely predicts tonight's. The variable jumps. A supplement moving 2 minutes works on a variable that naturally swings 10 or 12 minutes between nights.

That night-to-night independence means the compound's effect must be consistently large to emerge from the noise floor. Two minutes does not reach that scale, regardless of the follow-up horizon the model covers.

The subjective-objective gap as data

72% active versus 29% placebo. That is a strong signal. It is not noise. It is 42 percentage points on a binary question: did you sleep better?

The actigraphy recorded 2 minutes.

Neither reading cancels the other. The gap documents the distance between what the body reports and what the accelerometer captures. Subjective sleep quality and actigraphy latency measure different things.

That gap is information. It is not a reason to dismiss the trial. It is also not a reason to present the trial as proof of efficacy.

The instrument's position

The ring records nights, not feelings. It sees the latency the algorithm computes from movement data. It cannot see your perception of rest.

Across 650 nights of case zero, sleep latency had a coefficient of variation of 84.41%. Across the maximum 712 nights in the dataset, that number does not drop enough to make -16.7% detectable.

For an honest verdict, the effect must exceed 18.53% at 168 nights. -16.7% does not reach it.

The limit

This data says the published effect in 150 healthy adults does not cross our detection threshold at any horizon up to 168 nights. It does not say ashwagandha is ineffective. It says nothing about the subjective signal, which the trial reported as large. It does not say what would happen with a subject whose baseline latency differs from case zero's. It does not address a longer follow-up.

Case zero is one body. The thresholds come from 650 nights of that body. A different body with different baseline variability would have different thresholds.

The question this piece does not answer: what threshold would a subject with a 6-minute baseline latency — and less inherent variability — actually face?

---

Mental Work — all protocols · related file: magnesium-for-sleep · open science — the dataset