Each cluster node after processing a message/transaction from the log, could publish the result of processing back as a message to the cluster log.

A separate clustered service running in the same cluster could look for all the results, check if any of the result is different from the others and determine divergence. We would know exactly which transaction caused divergence.

Once divergence is detected, would have to check if Aeron permits reconfiguring the cluster dynamically by removing the divergent node from the cluster.