Out of the Crisis · W. Edwards Deming

Reacting to the normal variation of a process as though it were a signal makes the process worse; most errors come from the system and not from people.

The idea

Deming distinguishes between the variation inherent in a stable process and the variation that indicates something has changed. Confusing the two and acting on the first — what he calls tampering — degrades the result. And he argues that most failures are attributable to the system, not to the people working inside it.

Why it matters to us

We build systems that produce numbers and trigger actions from them. A badly placed threshold on a noisy metric is exactly the tampering Deming describes, now carried out at greater speed and with less deliberation.

What we question

The book is about manufacturing processes with high volumes and fast cycles. Many of the decisions we support are infrequent, with little data and very uneven consequences, where statistical control contributes less than it appears to.

What changed

Before proposing an automatic alert, we check how much variation the metric shows when nothing is happening. Several alerts we had planned were never built after that check.

What it connects to

It connects to the question of how to communicate uncertainty, to the question about confidence metrics, and to the capability of evaluation, observability and control.