One could have sort of “permission rule” messages to be interpreted by clients to decide, during “replaying” a group with its messages, which action was valid and which wasn’t, and what can this specific user do to the group at this specific moment.

clients could then be overwhelmed with mass invalid messages by bad actors

There could be different types of messages, perhaps with references to “interpreter” messages containing scripts.

I don’t know how safe could that be, but deltachat does something like that