This may be a bug. I’ll have to check, but a separate push and pull replication may get the same checkpoint ID, which would explain the problem. If that’s true, I’ll file a bug report.
The CBL 2 replication protocol can push and pull over a single socket. So it’s assumed that you will run a single push/pull replication, rather than two replications. Is there a reason you need to run these separately? If not, you can fix the problem by switching to a single replicator.