How much does entreno sesiones have to change before it stops being noise? The threshold from 383 nights

For a change in training session count to clear the background noise and register as a real signal, it has to exceed 43.88% relative to the 28-night mean. That number doesn't come from a literature consensus or a group average: it comes from 383 nights of continuous tracking in a single body, measuring how much this variable moves when nothing important has changed.

With a 56-night reference window, the threshold drops to 31.03%. With 168 nights, it reaches 17.91%. The window you choose matters because it defines how much context you have to separate signal from ordinary fluctuation.

What the coefficient of variation reveals

entreno_sesiones has a coefficient of variation (CV) of 62.78%. CV measures how much a variable disperses around its central value, expressed as a percentage of that value. The higher the CV, the more the variable moves on its own — without any meaningful change driving it.

A CV of 62.78% is high. It means training session count jumps around considerably from week to week in the normal course of life: weeks with 4 sessions, weeks with 1, weeks with none. Travel, demanding work periods, recovery phases, difficult stretches — all of it produces oscillations that belong to the normal pattern, not to any signal worth tracking.

For something to register as a signal over that background, it has to clear that cloud. A change that falls short of the threshold stays inside the historical oscillation and can't be read as meaningful. The 43.88% threshold for 28 nights is the point where the probability of observing ordinary variation drops enough for a movement to count as something.

Short memory: what autocorrelation tells you

The second number that defines the character of entreno_sesiones is autocorrelation: rho = 0.281.

High autocorrelation — near 1 — would mean the variable builds strong inertia: a high-session week predicts another high-session week. Past behavior predicts future behavior well. Low autocorrelation — near 0 — means each cycle starts roughly from scratch. Whatever happened last week has little predictive power over what happens next.

At rho = 0.281, entreno_sesiones has short memory. Training sessions in this body, over this period, don't build clear momentum from cycle to cycle. A high-load week doesn't reliably predict the following week. That amplifies the need for conservative thresholds: when a variable is both volatile and low-memory, individual movements say even less.

The 383 nights: full table from the source row

The dataset covers 383 nights of a single subject, tracked alongside Oura ring data. The session median is 2 — the point where half of all cycles fell below and half fell above. Median is more robust than mean for variables with this level of dispersion.

Metric	Value
Nights tracked	383
Session median	2
Coefficient of variation (CV)	62.78%
Autocorrelation (rho)	0.281
Relative CV (rcv)	174.02%
Threshold at 28 nights	43.88%
Threshold at 56 nights	31.03%
Threshold at 84 nights	25.33%
Threshold at 168 nights	17.91%

The relative coefficient of variation (rcv) of 174.02% is the ratio of CV to median. With a median of 2 and a CV of 62.78%, a variation of one full session represents an enormous fraction of the central value. The rcv of 174.02% is the compressed way of saying this variable is extremely difficult to read at fine granularity.

The direction in the thresholds is consistent: more reference nights, lower threshold. With 168 nights of baseline, the system has enough history to separate signal from noise with more precision. With 28 nights, the base is smaller and the threshold has to be more conservative to avoid false reads.

Reading the thresholds in practice

Thresholds aren't action rules: they're interpretation tools. They tell you when a movement in session count can be taken seriously as information about the system, and when it's more likely to be background variation.

An example: your 28-night mean is 2 sessions and this week you logged 4. That change clears the 43.88% threshold — the system registers a real shift. What the shift means depends on context: a planned high-load phase, a stress response, the beginning of a load cycle worth tracking.

What the 43.88% threshold reveals with a median of 2 is that, at integer session resolution, almost any single-step movement already crosses the line. The signal/noise problem for this variable isn't a matter of degree: it's structural. The variable is too volatile at weekly resolution for fine-grained signals.

Where this variable fits in the full system

The system covers 31 variables and 18483 measurements across 650 nights total. entreno_sesiones ranks among the highest-variability variables in the set. Biometric variables — resting heart rate, heart rate variability, respiratory rate — have more compact thresholds because physiology regulates them much more tightly. Behavior has more degrees of freedom than biology.

That doesn't make entreno_sesiones less valuable: it makes it different. Its value isn't in fine-grained weekly tracking; it's in detecting extended periods of abandonment or structurally high load. An rcv of 174.02% places this variable in the category of coarse context, not fine signal.

The limit

This threshold is the signature of a single body across 383 nights. The 43.88% for 28 nights is not a human standard or an external reference: it's the result of measuring how much this variable moves in this life, with this protocol, over this period of time.

The CV can change. If the subject adopts a more stable and consistent training routine over the coming months, the CV falls, the dispersion compresses, and the threshold falls with it. The threshold isn't fixed: it's a function of the accumulated record and gets recalculated as the file grows.

This data doesn't say whether training sessions are associated with any biometric variable. That's a different question, with its own correlation analysis. The only question answered here is how much entreno_sesiones has to move for that movement to be distinguishable from its own historical variability.

No comparison with another subject is made because the protocol doesn't allow it. The house apparatus rule establishes that absolute values from two subjects measured with different instruments can't be compared. What can be compared across subjects is each one's relative variability against themselves.

---

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