Some of the actions are automated based on some NN algorithm score, and then the appeals are human-powered. They have large third party content review offices that are operated like call centers in which humans review these things. I understand they're real meat grinders to work in.

I've reported clearly racist, harassing content before and had the reviewer report it as confirming to their standards. I know people that were banned for bullying for wishing people happy birthday. As much as I suspected a bunch of people are just quickly mashing random buttons to pump up their score, I read that they're evaluated based on the success and failure of appeals to their judgements, so I can't imagine they would be. There are clearly deep-seated problems with this process.