Every change has a cost
Significant ones restart learning. During that window the numbers are unreliable, which prompts more changes, which extends the window.
What to keep doing continuously
Negatives, budget shifts between proven campaigns, clearing disapprovals, pausing what is clearly bleeding. None of these meaningfully disturb learning.
What to do rarely and deliberately
Structure, bid strategy, targets, conversion actions. One at a time, spaced far enough apart to read the result.
An account that looks calm in its change history is usually being managed better than one that looks busy.