This is ultra-low-priority junk I had stuck in my head last night, and I figured writing it all down might be the best way to stop thinking about it.
One of the issues with Sparkplug B is support for systems with multiple application nodes. Specifically, in terms of coordinating BIRTH messages.
These thoughts are not compatible with Sparkplug B, but are still interesting, possibly for a future generation... I'm trying to think them through before badgering people on the committee with them.
I'm contemplating what happens if you split the variable and non-variable portions of the NBIRTH into two messages, publishing the non-variant portion persistently, and using the LWT to both announce the node has gone offline and clear the persistent publish.
Let's call the new persistent topic NSCHEMA.