Thanks, that makes more sense now with the example.
I think given the bespoke logic that you’re looking at here will require either a customer Single Message Transform, or a Kafka Streams/ksqlDB stream processing application.
There is the Filter SMT from Confluent (not the same as the Apache Kafka Filter) which can be used to conditionally apply Single Message Transform, but only to drop/include a record, not conditionally apply another SMT.