personally i think there should be a "replace" but keep the originals, that would be more useful than delete

in distributed databases they often have a thing called a "tombstone" which "marks" an item deleted and the reason being that you can't guarantee all replicas are going to delete it because it's decentralised

delete events can be treated the same way, but the guys who invented this protocol don't understand concurrency and race conditions, or at least, they are starting to understand them now, after building this whole thing that is inherently highly concurrent