The cost of a fast feedback loop

Everyone agrees CI should be fast. Almost nobody budgets for it. What follows is a concrete accounting of what we spent making the loop quick, and what that money bought back in attention that would otherwise have leaked away between pushing a branch and learning whether it worked.

The interesting number is not the build time. It is the number of times a person context-switches while waiting, because every one of those switches costs more than the minutes on the clock suggest.