Rebuttals take much longer to write.
Assuming that two endpoints are using the same or forward-compatible schemas, there's no "assumption" involved: if you sent a message of type T, it gets decoded as type T. There isn't protection against inauthentic messages, admittedly.
Would you like to cite an RPC scheme that meets your definition of type safety.