You could create an additional recorded IPC publication from within the clustered service that records all messages that are outbound (let's call this the egress-log). A separate application/process can then read the recorded egress log of each cluster node and compare the messages to detect divergence. To help manage the comparison of messages by this "divergence detecter" process I would have the cluster stamp each message with a monotonically increasing sequence number so you can ensure you properly comparing the same message (as well as detect gaps/dupes).