Main idea

Some measures genuinely shed light: rework, cycle time, regressions, defects found before production, defects found in production, the support load tied to certain defects. They tell you whether the team redoes its work too often, whether flows break regularly, whether one area of the product generates too much support, whether detection comes too late.

They stop shedding light the moment the time spent qualifying exceeds the time spent improving. The queue of defects has then been replaced by a queue of categories, with the same effect: sustained activity standing in for treatment.

Layer added by "Quality belongs to those who ship" (2026-06-03). A simpler criterion comes before the one about time spent, and it can be tested on each qualification taken in isolation: does this qualification change a decision? Arguing over whether an item is a defect, a feature, an improvement, a rework or a QA anomaly costs nothing as long as the outcome changes what will be done. When it doesn't, the time spent is already lost, whatever the size of the measurement apparatus. Measurement sheds light on a decision; it does not replace it.

Why it matters

The exit criterion is not how fine-grained the measurement is — a fine-grained measure can be perfectly useful. It is its relation to the corrective work it is supposed to steer.

Nuances and limits

An instrumentation phase temporarily costs more than it returns, without that being a drift.

And the comparison between the two amounts of time isn't always measurable; it often remains a team judgment.

Open questions

• Which indicator signals that measurement has become the work rather than its instrument?