With a centralized service, trust is simple: how much you trust the single entity that represents the service.

In a distributed network, nodes need to build trust to each other. In the best-known federated network, email, domain reputation is a thing. Various blacklists and graylists pass around trust values in bulk.

So a node with a ton of sock puppets trying to spam votes (or content) is going to lose the trust of its peers fast, so the spam from it will end up marked as such. A well-run node will gain considerable trust with time.

This, of course, while helpful, does not guarantee "fairness" of any kind. If technology and people's values clash, the values prevail. You cannot alter values with technology alone (even weapon technology).