@Jan van Brügge Jup, pretty much. The mechanism is fully general though, it is not tied to any decision about actually want to provide replication-by-type in the core or not. That's a decision for (slightly) higher up in the protocol stack.

The merkle-ooo approach is indeed strictly more powerful, but also strictly more complex. The different sigchains are much less expressive, but they wouldn't require any additional metadata overhead.