The breakdown of a project into milestones of predictable difficulty generally has high error bars. If something has been built before then you can predict it but why are you building it again?
The only way I've found of doing this accurately is to reassign a task, IE if something is taking dev1 what you perceive as too long, assign it to less experienced dev2 and see if they can complete it from scratch in less time. If so, you know dev1 is subpar or slacking and should be RIF'd.
The milestones can iteratively evolve, so can the timeline. Breaking a project into milestones is a highly simplified example. The idea is that each step of the way we should be able to hold engineers to the same bar with or without return-to-office.
The only way I've found of doing this accurately is to reassign a task, IE if something is taking dev1 what you perceive as too long, assign it to less experienced dev2 and see if they can complete it from scratch in less time. If so, you know dev1 is subpar or slacking and should be RIF'd.