Part VI — What the answer rests on

Given a wide interval, what is the one thing to go and do about it?

An interval describes a problem without saying what to do about it. This part answers that, and then shows the limit of everything the book has built.

ch19 · Which input to go and measure ranks the inputs by how far each one moves the answer. Then it answers the question a ranking raises: what measuring the top one would buy, as an upper bound. Two of its results would carry the chapter on their own. The book’s most carefully measured constants buy nothing, and in the observability model the input that ties for first cannot be measured at all.

ch20 · The missing node is the error no technique in this book can rank, because it is not in the list. A model can be wrong about a number or wrong about its shape. Everything in Parts IV and VI is about the first. The second is where the machinery stops helping and starts flattering.

The order matters. A reader who does ch19 without ch20 has learned to optimise a list without asking whether the list is complete.