When a slave HC registers, the master provides it a copy of the domain-wide model as a DMR tree. This has a weakness in that the Resource impl types that comprise that tree on the master side are not transmitted. If custom Resource impls are used, they may not be recreated on the slave.
Consider instead sending a list of mgmt ops, a la what we do with servers when starting from an HC.
This could potentially be limited to subsystems, with the core model sent as it is now. The core model is handled by core-AS code on the slave side, so we can reasonably ensure that code always uses the correct Resource type. There are places where we already do that. The bigger problem is subsystems.