I am optimistic that a social graph can help with this, using transitive trust and making it easy to transitively "cut off" bad parts of the graph with a blockl

I am optimistic that a social graph can help with this, using transitive trust and making it easy to transitively "cut off" bad parts of the graph with a blocklist. The problem is that commercial graphs do become useless, as you say. But with the Fediverse we now have a graph that is resistant to this. I'm not aware of anyone building a reputation system on top of it, but this is something I expect will develop in the future. The system's opinion about a particular data point would then be personalized based on the people you say you trust to indicate that information, combined with the people they say they trust, and so forth.