A stream filter evaluates each event and forwards only matching records. It does not change the source log or turn an unmatched record into a deletion. Business conditions matter: a canceled order can still carry a positive amount, so amount alone does not establish paid revenue. These tasks filter fictional order events by their status field and preserve the matching key and value. The local stream control models the result of a topology rather than compiling Kafka Streams Java code. Inspect the output log to check which event survived. Counting one output is not enough if it came from the wrong input. A correct filter preserves the meaning of the requested condition.
Inspect the topics, partition logs and group offsets before changing the cluster.