Interesting, if that's so, can we redefine the underlying Kafka topics as "backups" and achieve compliance by having the stream processors drop "forgotten" records when replaying a topic?