How much does met alto min have to change before it stops being noise? The threshold from 712 nights

For a change in met alto min — the minutes a body spends in high-MET activity overnight — to count as more than noise, it has to move by more than 72.6% when you only have 28 nights of history behind you. That threshold drops to 51.34% at 56 nights, to 41.92% at 84 nights, and to 29.64% at 168 nights. This is the number that separates a real shift from a fluctuation the variable produces entirely on its own, night after night, with nothing behind it.

What met alto min is, and why its noise is so large

met_alto_min counts minutes of high-MET activity captured during the overnight window. In the body measured here — 712 nights, one subject, one device — the median of the series is 0. On most nights the value is literally zero: there are no high-MET minutes to log at all. That fact alone tells you something before you look at any other number: this is a rare-event variable, not a continuous trend like resting heart rate or heart-rate variability.

When a series has a median of 0, the coefficient of variation explodes, because any single night with a nonzero value dominates the calculation. Here the coefficient of variation is 155.02%. Compare that to a stable variable, where the mean and standard deviation sit at similar orders of magnitude and the CV stays below the average itself. At 155.02%, this variable swings harder than its own average — the statistical signature of an intermittent event, not a steady biological rhythm.

The day-to-day correlation (rho) is -0.112. In plain terms: there is almost no memory from one night to the next. Whether you log high-MET minutes tonight tells you almost nothing about whether you will tomorrow. A rho this close to zero, and slightly negative, is the same story told a second way — this variable does not carry a trend forward. It appears and disappears.

How the threshold was built (and why it shifts with the window)

The noise threshold answers one question: if you take a stretch of N consecutive nights, how much does the average of met_alto_min typically vary from one stretch to the next, by chance alone? That stretch-to-stretch swing is the rcv — the coefficient of variation of the windowed reports. Here the rcv is 429.69%. It is enormous for the same reason as before: because the typical value sits at zero, any stretch that includes a few high-MET minutes looks like a massive jump against a stretch that reads zero.

The window thresholds fall out of that rcv. With fewer accumulated nights (28), the threshold is higher — 72.6% — because there is less data available to average the natural noise down. As the window grows, the average settles and the threshold drops: 51.34% at 56 nights, 41.92% at 84 nights, 29.64% at 168 nights. The right way to read this is not "the data gets better with time" — it is that you need more accumulated nights before a change in met_alto_min stops looking like what the variable does on its own, with no identifiable cause behind it.

Window (nights)	Noise threshold
28	72.6%
56	51.34%
84	41.92%
168	29.64%

Full-series metric	Value
Nights measured	712
Median	0
Coefficient of variation (CV)	155.02%
Day-to-day correlation (rho)	-0.112
Windowed-report CV (rcv)	429.69%

This table comes straight from the ruido.medido.json#met_alto_min row, one of 31 variables in a file built on 712 nights from a single body. No rounding, no interpolation — every figure here traces back to that row.

The method behind this calculation is not unique to met_alto_min — it applies the same way to any variable in the series, with the window size adjusted to that variable's own behavior. What changes from one variable to the next is how close its typical value sits to zero, and how little memory carries from one night to the next — two factors that together decide whether background noise stays manageable or ends up dominating the daily reading entirely. For rare-event variables like this one, ignoring the background noise means reading meaning into what is really just chance: every spike is associated with a signal when it is actually the expected behavior of an intermittent event that appears and disappears without carrying anything forward from one night to the next.

What this means for reading your own number

If you see met_alto_min go up on a given day, the right question is not "did it go up" but "did it go up enough to stop looking like what this variable does by itself?" With little accumulated history (28 nights), you need a change of more than 72.6% before you should even suspect something other than noise is behind it. With more history (168 nights), that bar drops to 29.64% — still a large change, just less brutal than it was at the start.

This is not a quirk of one unusual body. It is the mathematical signature of any variable whose natural median sits at 0. Most sleep-tracking tools show you the day's number without ever telling you what its background noise looks like. Here, the background noise is the headline finding, not a footnote.

This also explains why comparing one single day against another single day almost never works for this particular variable. The near-total absence of day-to-day memory means yesterday's value anchors no reasonable expectation about today's. What actually works is comparing full windows against each other — one stretch of several nights against another stretch of similar size — because only then does the average get a chance to settle before it's asked to say anything.

The limit

This is one person, one device, 712 nights. The threshold describes the statistical behavior of this series in this body — it is not a universal constant you can apply to another subject or another sensor without measuring again. The 712-night, 31-metric figure sits inside the same file as 18483 measurements overall, but met_alto_min does not share its variance with the others: every variable in that file carries its own threshold because each one behaves differently. This data also does not say what caused a given night's high-MET minutes, or whether that minute matters for anything in your waking life. It only marks the point where a change stops resembling the noise the variable makes on its own. It is not a verdict about your health or about any practice — it is the ruler you need before any verdict.

Keep reading

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