Katballe Studio Writing Projects About
June 1, 2026

Measuring Slow Systems

AbstractA short method note on measuring systems whose effects unfold over long timescales — where the signal you care about is easily drowned by noise that arrives sooner. This is a sample whitepaper showing the abstract block, an auto-generated table of contents, and footnotes. Delete it whenever you like.

Some systems reveal themselves quickly; others take their time. The hard part of measuring the slow ones is not the measurement — it is the patience and the bookkeeping required to tell signal from noise.1

The problem with fast proxies

When the outcome we care about is slow, we reach for a faster proxy. Proxies are seductive because they move now. But a proxy is only as good as its correlation with the real outcome, and that correlation is rarely stable.

Goodhart's shadow

Once a proxy becomes a target, people optimize the proxy rather than the outcome.2 The measure stops measuring. The defense is to keep at least one slow, expensive, honest metric that nobody is allowed to game.

Designing the assay

A good assay for a slow system has three properties:

  1. A stable baseline — you cannot detect drift without it.
  2. A unit of observation that survives aggregation.
  3. A pre-registered question, decided before the data arrives.

What to do while you wait

Waiting is not idleness. Instrument the system, write down predictions, and let the slow metric accumulate. The discipline is to not act on the fast proxy before the slow signal has had time to speak.

Footnotes

  1. Noise that arrives sooner is not more important — only louder. ↩

  2. This is Goodhart's law, paraphrased: when a measure becomes a target, it ceases to be a good measure. ↩