Worth being precise here: what the label says and what the trial protocols permitted are different documents, and the difference is instructive: protocols generally allowed a step to be delayed or reversed for intolerance, and a substantial minority of participants used that provision.
Escalating in response to a plateau is a specific and common error of reasoning. A plateau after four to six months is the expected trajectory in every trial in the class — the curve flattens because energy expenditure falls with mass, not because the receptor stopped working. A dose step may still be reasonable; "the loss stopped" is not by itself the reason.
In practice, missing a dose by three days on a weekly schedule takes the trough down by less than a factor of 1.35, which is well inside the variation you would see between two on-time weeks. Missing it by five or more days is where the label instructions start to differ between agents, and the reason is how close the next scheduled dose is rather than any mechanistic threshold.
SURMOUNT-4 is the withdrawal trial to read on the maintenance question: after an open-label lead-in, randomised withdrawal produced substantial regain in the placebo arm while continued treatment produced continued loss. It is the cleanest available answer to "what happens if I stop".
Escalate on tolerability, hold when it costs you, and do not confuse a flattening curve with a failing drug.