How many nights does glycine need before you can judge it (measured on sleep latency)?

The number is 140 nights. Not 7. A 2016 crossover trial found that 3 g of free-form glycine taken 30–60 minutes before bed reduced sleep latency by -20.7% — from 13.5 min to 10.7 min — over 7 days. That -20.7% relative effect just clears the detection threshold that 650 nights of our own data set for 140 nights: 20.29%. Before that mark, your body's inherent noise in this variable covers the effect. No reading is valid.

The finding

A 2016 crossover trial enrolled 11 adults who self-reported insomnia. The protocol: 3 g of free-form glycine, taken 30–60 minutes before sleep, for 7 days. The measured outcome was sleep onset latency. Starting mean: 13.5 min. Ending mean: 10.7 min. That is a -2.8 min absolute change, or -20.7% in relative terms.

The design was a small crossover (n = 11) with a specific population: adults who subjectively struggled to fall asleep. The trial was not conducted in healthy adults without a baseline sleep difficulty. The source row registers this explicitly: transfer to a healthy adult is not established.

Parameter	Value
Compound	free-form glycine (amino acid)
Dose	3 g, 30–60 min before bed
Duration	7 days
Population	11 adults, self-reported insomnia
Initial latency (mean)	13.5 min
Final latency (mean)	10.7 min
Absolute change	-2.8 min
Relative change	-20.7%
Design	crossover RCT
Reference year	2016

Source row: catalogo.generado.json#evidencia[3]

The gap (H3)

The trial measures a group-level effect. What it does not measure is how much noise any individual's sleep latency carries over time — and whether a -20.7% shift is distinguishable from that noise at 7 days, 28, 56, or 140. Without a personal variability baseline, the literature figure cannot be transported to your case. There is no way to know whether a change from one night to the next is signal or fluctuation.

This is gap H3: an effect documented at the group level with no translation to the individual.

The transfer

The bottle on your nightstand carries the same gap. Seven days is not enough to judge whether glycine works for you, because no reference floor exists to tell you when a change stops being noise and starts being a real effect.

Sleep latency is a high-noise variable. In the 650-night case-zero dataset, the coefficient of variation for sleep_latency_min is 84.41%. Night-to-night latency swings substantially, and yesterday's number does not predict tonight's: the serial correlation coefficient (rho) is 0.026 — near zero. With that level of internal variability, 7 nights produce no readable signal. The question is not whether glycine did something during that window. The question is whether the data window is wide enough to answer either way.

The instrument: the 140-night threshold

The case-zero dataset spans 650 nights and 18483 measurements across 31 tracked metrics. For sleep_latency_min, the case-zero median is 11 min. The relative coefficient of variation (RCV) for this variable reaches 233.96% — one of the highest in the dataset. With that level of noise, the detectable-change threshold falls only slowly as nights accumulate.

Nights accumulated	Detectable change threshold
28	45.38%
56	32.09%
84	26.2%
112	22.69%
140	20.29%
168	18.53%

Source rows: ruido.medido.json#sleep_latency_min · catalogo.generado.json#evidencia[3]

The published effect is -20.7%. The 140-night threshold is 20.29%. They are nearly identical. Reading: if the effect exists in your body at the same level as in the literature, you need exactly 140 nights of clean data to separate that signal from your latency noise.

At 28 nights the threshold is 45.38%: a -20.7% effect stays buried. At 56 it drops to 32.09%, still out of range. At 84 it reaches 26.2%, and at 112 it falls to 22.69%. Only at 140 nights does the threshold (20.29%) cross below the published effect (-20.7%). From that point forward — and not before — a change of that magnitude becomes readable above noise.

Three statistics drive this threshold. The serial correlation (rho = 0.026) confirms that sleep latency is essentially random from night to night — there is no momentum to exploit. The coefficient of variation (84.41%) marks how wide the nightly spread is. And the RCV (233.96%) shows how much that spread exceeds what most variables carry. These three numbers are why the answer is 140 and not 7.

The limit

Four things this data does not say:

1. It does not confirm glycine works in healthy adults. The 2016 trial used 11 adults with self-reported insomnia. Transfer to an adult with no baseline sleep-onset difficulty is not established. The source row flags this explicitly: transfiere_a_sano: false. If your sleep latency is already 11 min or under in normal conditions, the trial's starting point does not match yours.

2. The primary citation is unverified. The -20.7% effect is referenced as a "2016 crossover trial," but the DOI was not located in this run. This figure comes from a secondary source without PubMed verification. The archive marks it as no_verificado. If and when the primary citation is confirmed, this cell will be updated.

3. The absolute change falls below the individual detection threshold. The -2.8 min absolute shift is below the 6.5 min minimum detectable change for this instrument at the individual level. What crosses the threshold at 140 nights is the relative effect (-20.7% vs 20.29%). In real minutes, the trial's absolute effect is not detectable in a single subject using this method.

4. The threshold belongs to the case-zero subject, not to you. The 650 nights come from one person measured with an Oura ring over years. Your intrinsic variability may differ. The 20.29% threshold at 140 nights is a dataset reference, not a universal standard.

The inherited gap

This file closes the question "how many nights?" for sleep latency and glycine. It does not answer the next question: what does it mean if after 140 nights the change does not reach 20.29%? Did glycine not work — or does an effect exist below the instrument's floor? That belongs to the expedition on the null as a result, not this file.

---

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