The Cost of Keeping Score

For all of Q2 2026 I timed my own measuring. Every weigh-in, every log review, every drill, every spreadsheet I opened in service of knowing a number about my own life, recorded to the nearest minute.

This was uncomfortable to do and the result is uncomfortable to publish, which is usually a sign the measurement was worth taking. One of my practices costs fifteen and a half hours per decision it has ever changed.

The ledger

Practice	Minutes per month	Decisions changed, Q2
Food self-provisioning weigh-ins	310	1
Backup restore drills	55	1
Sovereignty score computation	38	2
Media and subscription audit	26	3
Self-custody drills	22	0
Smart home DNS log review	18	1
Energy logging (automated)	4	0
Subtotal, measurement	473	8
Writing this log	340	n/a
Total	813

Eight hundred and thirteen minutes a month. 13.6 hours. Just under 8 hours excluding the writing, which belongs in a different category since it is the output rather than the instrument.

Eight hours a month is a part-time hobby. I had been carrying it as a background process, the way people carry commuting, and the number is much larger than the impression.

Cost per decision changed

The right denominator is not minutes and it is not insights. It is decisions I made differently because of the data. A measurement that has never changed an action is a measurement I am performing for the feeling of rigor, and I would rather catch myself doing that than have a reader catch me.

Food weigh-ins: 310 minutes per month, one decision. That is 930 minutes for one decision, fifteen and a half hours. The decision was worth making. It was not worth fifteen hours.

Backup restore drills: 55 minutes per month, one decision. That decision was discovering the offline drive was 41 days stale while my dashboard was green. Call it 165 minutes for a finding that would otherwise have surfaced during an actual incident, at the worst possible moment. This one I would pay ten times over.

Self-custody drills: 22 minutes per month, zero decisions. Zero looks damning and is not. The drill keeps a capability current for an event that has not happened. Its value is optionality, not information, and the cost-per-decision framing does not apply. I flag this because the framing is a tool, and tools misapplied produce confident nonsense. Insurance is not a failed investment.

Energy logging: 4 minutes per month. Fully automated, effectively free, keep forever. Automation is what a measurement should graduate into.

The cut

The food weigh-ins get retired in their daily form.

This is the June entry where I weighed every gram consumed for four weeks to compute a self-provisioning ratio. It was genuinely revealing the first time. What I never questioned was continuing it indefinitely, which happened by inertia rather than by choice.

Before cutting I checked whether the signal survives sampling. I took last year's full-quarter data and recomputed the ratio from a two-week window, repeated for every possible window. The sampled estimate landed within 3.1% of the full-quarter figure, with the worst window off by 5.4% during a period containing a holiday.

So: a two-week sample per quarter instead of continuous logging. 310 minutes a month becomes about 52. I lose roughly 3% of accuracy on a metric that moves 10 to 15 points a year. Not a close call, and the only reason I did not make it eighteen months ago is that I never priced the practice.

New total: 555 minutes a month, of which 340 is writing. Measurement proper drops from 473 minutes to 215.

The rule

A measurement earns its place if it can change a decision or preserve a capability. If it can do neither, it is a number I keep because keeping numbers feels like diligence.

Applied honestly this is a stricter filter than it sounds. It would have killed the food weigh-ins a year earlier. It would have caught, much sooner, that I was reviewing smart home DNS logs by hand every week when a threshold alert does the same job in 4 minutes a month. It does not touch the drills, because "preserve a capability" is doing real work in that sentence.

The part that applies to everyone else

The measurement stack is itself a dependency, and I have spent this month auditing dependencies without once auditing this one. It has a cost, it accumulates by inertia, it is defended by identity rather than by evidence, and nothing in the system prompts you to review it. That is the exact profile of every subscription I cancelled in the media audit, and I did not notice the shape until I put it in a table.

The whole doctrine of this site is that you cannot own what you have not measured. The correction I am adding, at some expense to my own self-image: you cannot sustain what you have not priced. A sovereignty practice that consumes eight hours a month will be quietly abandoned within two years, and abandoned practices protect nobody. The version that survives is the version that costs little enough that you keep doing it on the week when everything else goes wrong.

Automate what you can. Sample what you cannot. Retire what has never changed your mind. Keep the drills.

Confidence 0.89 that the sampled food metric stays within 5% of the continuous figure over the next four quarters, which is falsifiable, and which I will check in July 2027 against a full-month control.

Measure everything. Then measure the measuring.