You see, the thing is that there are multiple reasons why recognition may not work properly. Face could be too uncertain due to low lighting of the photo, or too low resolution, or person might be looking away, or that might be not a face at all, but incorrectly identified as face (face of a statue or face on a painting, for example). Also, face can be considered as recognised, but it can be assigned to wrong person. Or there could even be no face detected on the photo, while there is.
My point is, if AI would have means to detect its own mistakes, it would not do those mistakes in the first place. So it cannot really help in detecting where it went wrong, or what it has missed, unfortunately.
If the goal is to have 100% faces marked up, I guess the only reasonable way is just doing a full review of all photos. Not that I would suggest doing that, but if that’s the goal, then first you need to decide the order in which you would go thru all of your collection. Will it be folder by folder or event by event on the time line. Or, maybe there is just limited set of folders or events - that would simplify the task.
Then I would suggest going to the first photo of next folder/event, open it then, open File Info panel and then just click next, next, next. File info panel will give immediate view on all identified people on the image, as well as detected, but unassigned faces. Also, you could quickly see if the number of faces matches number of people on the photo (if that is not a group photo). And take immediate action in case of mistakes.
For a group photos you can’t immediately tell if everything is OK just by looking to the File Info panel. Here you can use another trick - hover your mouse over ever single face on the image and check if all of them were identified, if they were tagged properly. This way you could do it quite efficiently.
However, my recommendation would be making pauses in that process once you made significant amount of manual changes, switching back to People mode, and see what AI has to suggest, based on the new information you have given it to learn on. Since this process is 1000 times more efficient, you don’t wont to go back to manual review again until AI is already done with its suggestions for now.
Hope that helps.