What an instrument is for

Every engineering organisation is full of instruments. Dashboards, architecture diagrams, burn-down charts, risk matrices. They all make the same implicit promise: lean in, and you will find something that was true before you looked.

Most of them break that promise. The architecture diagram shows boundaries that stopped existing two refactors ago. The dashboard shows a number nobody can trace to a query. The risk matrix shows colours that were chosen in a meeting, by feel. Each of these is a drawing of a dial — the shape of measurement without the measurement.

The difference matters because instruments are how organisations think. A team does not reason about its system directly; it reasons about the pictures it has of the system. When the pictures are decorative, the reasoning is decorative too, and the first real contact with reality happens in production, at the worst possible time.

So the test I apply to any instrument — a metric, a diagram, a status page — is whether it can lose an argument. A real instrument can show you a number you did not want. It can contradict the roadmap. If there is no way for the picture to embarrass its owner, it is not an instrument, it is marketing aimed inward.

Building instruments that can lose arguments is more expensive. The diagram has to be generated from the system, not from memory of it. The number has to have a query behind it that anyone can run. This is exactly the cost that makes them worth something: an instrument you cannot argue with is one nobody will bother to read.

The same discipline applies to sentences. A claim that cannot be wrong is not carrying anything either.