Section 71 of 440

PART 4: IMPLEMENTATION

4.8 MEASUREMENT INTEGRITY & EPISTEMIC HYGIENE

This book has told you, relentlessly, that measurement enables management. Here is the counterweight that keeps the whole system honest: measurement can also deceive management. A tracking system without epistemic hygiene does not make you scientific. It makes you confidently wrong with charts.

The Failure Modes

• Goodhart’s law. When a measure becomes a target, it stops being a good measure. Chase the sleep score and you will optimize the score — lying rigidly still, anxious about the number — rather than the sleep. Track behaviors you control; let outcomes follow.

• Instrument error. Consumer wearables disagree with laboratories and with each other. Never compare readings across devices, and treat any single reading as a rumor.

• Current result versus prior comparable result. Preserve dates, protocol, support, uncertainty and the detectable-change boundary; do not infer change from one point.

• Regression to the mean. Extreme days are usually followed by more ordinary ones with no intervention at all. Start a protocol on your worst week and it will "work" — because almost anything would have.

• False precision. A 3.74 is not more meaningful than a 3.7. Decimals from self-rated inputs are texture, not truth; the Engine rounds because honesty requires it.

• Correlation is not causation. Your energy rose after the new supplement — and after the days lengthened, the deadline passed, and you slept more. N-of-1 changes one variable at a time because otherwise the data cannot tell you which story is true.

• Expectation effects. In self-experiments you are subject, scientist, and cheerleader at once. Decide before you begin what would count as failure, and honor it.

The Hygiene Rules

Set a minimum observation window before judging any change — for most lifestyle variables, weeks, not days. Write a stop rule when you start ("if no trend change in six weeks, this ends") so sunk cost cannot vote later. Handle missing data honestly: a blank is a blank, never a zero, and the Engine’s completion gates enforce this so a skipped day cannot masquerade as a bad one. Know when not to track — during grief (section 4.2), when tracking itself breeds anxiety or disordered patterns around food and body, or when the ritual has replaced the behavior it was meant to serve. Stopping the count is sometimes the highest-integrity measurement decision available. And practice data minimisation: collect what you use, delete what you do not, and remember that a spreadsheet of your body is worth protecting like one.

THE PRINCIPLE: The point of measuring is fewer illusions, not more numbers. The moment your instruments start manufacturing illusions, recalibrate the instruments — or put them down.