One could start with a model similar to that in which most people using systems which this new system is meant to replace already find themselves.

You should be able to select peers (though, maybe you should be given some as well?) and "experts" to follow. You would only see their activity when you want to, but you could also participate in the larger network. As a practical matter of implementation, I think allowing groups to run segregated networks (ie. their own "node") which can later be joined to each other makes sense for getting actual adoption and for reducing development, infrastructure, and support costs.

I think there are likely many problems with such a crude model, but that is where we find ourselves it seems. Waiting to think up something better might be prudent, but that might be difficult without trying something out.