you are.right. flow is a.bad.word to use to describe it. but to quote guffey you can't actually push the shoulder back either.
what actually happens is this
we will start with a loaded round in the chamber. now that case is slightly shorter than the chamber we strive for 1 to 2 thousands. when fired at the case body/shoulder junction part of that shoulder becomes case wall. at the neck/shoulder junction part of that shoulder becomes neck. fireforming on a very small scale
now when you resize that case the events are reversed. at the case body/shoulder junction a small piece of case wall becomes shoulder. at the shoulder/neck junction a small piece of neck becomes shoulder.
over several cycles that shoulder brass migrates into the neck. you have to realize for just a moment at firing that brass is plastic the only direction it can move is towards the neck. that is why you have to trim.
Is there some common number of fire, size/resize cycles that it takes to produce donuts or is it like adult diapers and depends? How detrimental to accuracy is seating a bullet into the donut area?