From watch to number.

1. 1

   The watch never stops sampling

   Heart rate every few seconds, motion, sleep, and an HRV-derived 'stress' index, written continuously to raw FIT files on the device. This is the firehose: hundreds of thousands of readings a day, far too much to reason about directly.

2. 2

   A day is collapsed to a handful of numbers

   Each day's firehose is reduced to a small set of features: average and peak stress, the day's lowest Body Battery, resting heart rate, sleep-window stress. The 1,440 minutes become one row in a table: the unit every test actually runs on.

3. 3

   Some numbers are hand-built

   A few features don't come ready-made. The overnight stress variability that stands in for HRV is custom-extracted from the raw sleep-window samples, with a minimum-samples gate so a thin, poorly-recorded night drops out rather than quietly distorting the result.

4. 4

   Everything is read against my own baseline

   A raw value means little on its own; what matters is how far it sits from my personal normal. So signals are scored within-person, against my own recent baseline, never against a population figure. This is the 'relative, not absolute' rule, and it was learned the hard way when a population threshold mis-fit my small range. See the discipline that locks the rest.

5. 5

   The human side is one number a day

   Against all of that sits a single hand-logged number: the daily felt-state, 1 to 10. It's the ground truth the watch is tested against, and the crashes the whole study turns on are labelled from it. See how a crash gets labelled.

What the score and the watch share

The score and the watch are two measurements of the same body, not two independent witnesses. The watch reads physiology directly: heart rate, motion, derived autonomic state. The score is a brain-translation of how all of that feels. When they agree, they're agreeing about the same underlying body state through different machinery, not confirming each other. When they disagree, the disagreement is itself informative: the score can register a layer of mood or cognitive load the watch can't see, and the watch can pick up a body shift the felt sense hasn't yet noticed. Why they line up at all (and where the whole site's argument starts) is three things that have to be true.

None of these steps is neutral. An average can bury a five-minute spike; a baseline can be drawn the wrong way; a gate can drop a real night. Naming the pipeline is what lets the surface stay simple without hiding the machinery underneath it.

Every variable and its derivation, in the research repo: DATA_DICTIONARY.md ↗. The plain-language version is the data dictionary.