It should be clear by now that the ATProto "labels" (as in https://docs.bsky.app/blog/blueskys-moderation-architecture) approach doesn't work as the form of pseudo-decentralized moderation they were probably intended to be, otherwise not a single person would have been banned from Bluesky ever.

I'm pretty sure they ban you from the AppView, not only from the PDS. It would make no sense otherwise and they have said so. No one is guaranteed any right to stay in their AppView, as evidenced by the amount of checks and limits they put on the front door. The reasons for each specific ban do not matter. As long as you find a centralized point there will be always many "reasons": app stores, governments, advertisers, blackmailers, personal quibbles from moderators, whatever. Labelers can work well for small things, I agree, but I think the Bluesky designers misjudged their efficacy against the big things.

If you want good content moderation study how Reddit did it, the most advanced algorithms for moderation so far, with several pillars.