Estimates are not the problem

What happened

Yet another framework for better estimation made the rounds. Story points, t-shirts, reference classes, forecasting from throughput.

Why it matters

All of them assume the thing being estimated holds still. In most teams it does not.

The variance is not in how long the work takes. It is in how many times the work is redefined while in progress.

What to learn

Track one number for a month: how many times a ticket’s acceptance criteria changed after work started. Compare it to your estimate accuracy. The correlation will be uncomfortable.

What to try

For the next feature, write the non-goals before the goals and get them signed. Then measure the same number again.

Worth stealing

The one-page PRD. The non-goals section exists entirely to end the week-three argument that redefines the work.