I wonder if a good distinction is user initiated actions versus system initiated. If the user begins the action, the changes are immediate and buffered to the interface that appears next.
But when the system initiates it (eg. notifications, popups), then the prior interface remains active.
This is not the only distinction, but it is one of them, and I think that one is a good idea. Another distinction is the results of the user initiated action, of whether the result is expected or unexpected, and that distinction is not always so clear.
But when the system initiates it (eg. notifications, popups), then the prior interface remains active.
There's this paper studying this, and I think more work on it too.. https://dl.acm.org/doi/full/10.1145/3660338