Slower & more controlled federation is really an interesting problem to solve if you take reblogs and replies into account 🤔
Hard blocking a whole domain is easy to implement and reason about, you just reject everything that comes from that domain and never dereference remote stuff from there.
Allowlisting is also easy to implement and reason about cuz you reject everything *except* what's in your allowlist.
So I can see why Mastodon allows those two options, even though the first is tacitly discouraged and the second is downright difficult to enable properly.
I think the main reason slower federation is desirable is to prevent these two common fedi phenomenon:
1. Your boostable post ending up in places you don't want it to.
2. Unpleasant replies and harassment from randos.
So maybe a better way of approaching the issue would be to say that boostable posts can be boosted by people on your instance, and people on Trusted instances, so they can potentially be seen by people on those instances and instances they federate with, but they can't be boosted by people on instances with no established trust rating (that is, instances you've had no prior interactions with, no mutual follows, etc). And the same could be true of replies or indeed likes.
It seems like that would solve the worst of the issues, in a way that's relatively easy to reason about and explain 🤔
What are people's thoughts on this?
@f0x I was considering that as well. I think outgoing follows requests and lookups would be OK, and incoming follow requests and lookups as well. And then you can dereference boosts from people you follow, and that's how your instance dips its toes in.