Email spam filtering is already a lot more powerful, since it can consider things such as IP reputation. (IP reputation would be a non-starter for a federated social network, since it would require passing raw IPs along with messages in the federation protocol. And that is just unthinkable from a privacy perspective).
When all entities posting messages on a social network are managed by the same identity provider, and the identity provider can share information about the account with the social network, you get yet another layer of increase in power. You can, for example, tell whether the account is newly created or not. Or whether it was bulk-created along with 10 other accounts posting similar spam. Or a dozen other account reputation signals which let you classify accounts into organic vs. abusive, and use that to inform the classification of messages from those accounts. And again, the only way this happens is if the identity provider and the social network are one and the same. If they were separate companies, it'd be inexcusable for either party to share detailed information with the other.
"Hey, doing SMS verification of new accounts is expensive, but required by the other nodes on the network. How about we don't, but say we did?"
Imagine a solution to email spam that replaced looking at IP address reputation with looking at the sender's claim about their own reputation. Laughable idea, right?
The other problem with your suggestion is that you can't do abuse prevention with a single signal, even if it's not Boolean. I even gave an example in the previous post, with checking whether accounts doing similar abuse were clustered together at account creation time.
I guess nitpicking the details doesnt matter, but none of these problems seem insurmountable.
Once the biggest node on this network starts not peering with smaller nodes due to them being untrustworthy, what do you think happens? I suspect that the users will move from the smaller nodes to the bigger ones. And then we're right back to where we started. People will be complaining about monopolies and will want the big nodes broken up.
As for a solution, my intuition is that the first step might be to create a Sybil-proof voting system for determining how popular individual services are. If a service is refused federation by other services which collectively contain 90% of user accounts, then that's a sign that the service doesn't do enough to stop its users from posting objectionable content. Similarly, service A could report to service B how many abuse reports it is getting from its users about people on service B, with a clear rule about when service B would be blocked.
This isn't a perfect system, and I admit that a Sybil-proof global anonymous distributed voting system might be harder than just solving the Federator's Dilemma more directly, but I think it's worth considering, not least because such a voting system would be useful for countless other applications online.
Don't you have that already, and they have anti spam measures? How would a facebook post coming from facebook or twitter change what they remove as spam?