You can find a bunch of other threads, but the overly simple quickish version - the bubble screen and device tab are Sense screens where it is reporting using its ML learning process to identify devices. The simple explanation of how device detection works is pattern recognition based on what it sees power wise.
So it says " I saw a power signature of XXD happen which should mean device 3 is on. Lets display it as on". Then it does the same for the off. It has figured out in its detections that Device 3 turning off looks like YYY. So when it sees YYY happen in the power view, it knows to stop displaying Device 3 on the device views.
When you unplug the device manually, it may look YYZ instead of YYY because the shutdown process is different based on the device turning itself off and you yanking the plug from he wall. . So it misses the signal to say “oh, device 3 is off”. So the bubble stays on until a clean up process kicks in to realize that it must have made a mistake and turns off the device.
When you are in power graph view, you are looking at the “raw” readings from the CT clamps without any interpretation as to what that power means. So if you unplug a 500w thing, you’ll see that change on that screen right away since its a “Fact” that the power draw is now missing.
Does that make sense as to why the two screens can be off? One is “Fact”, one is interpreted data subject to mistakes.