Quick estimates that are good magnitude approximations but often somewhat wrong have a way of getting better, whereas estimates whose main purpose is to be with

Quick estimates that are good magnitude approximations but often somewhat wrong have a way of getting better, whereas estimates whose main purpose is to be within some delta of the actual time are working from the wrong value metric. A reasonably accurate aggregate of estimates doesn't require them all to have been perfected first. Teams that are obsessed with this will be slower to a build.