I like the idea of levels of trust, perhaps determined by algo and then manually adjustable.

To me you’re a third party, but given you’re just an npub and not amongst my friends of friends nor do i have any context about who you are, you’re not a trusted party to me. To others you might be. That’s the way it should work, it’s subjective. Any person or group should be able to provide a trusted third party moderation service.