During recovery eventsRead were not restored from the checkpoint and thus lead to a weird status after recovery, where we applied more events than we read:

❯ pcsm status
{
  "ok": true,
  "state": "running",
  "info": "Replicating Changes",
  "lagTimeSeconds": 2,
  "eventsRead": 322,     <======================
  "eventsApplied": 354, <======================
  "lastReplicatedOpTime": {
    "ts": "1763282460.1",
    "isoDate": "2025-11-16T08:41:00Z"
  },
  "initialSync": {
    "estimatedCloneSizeBytes": 12540,
    "clonedSizeBytes": 12540,
    "completed": true,
    "cloneCompleted": true
  }
}

That is because when we recover, we have eventsApplied correct and eventsRead set to zero.

Problem on top of this is also eventsRead value was also wrongly stored to the checkpoint, because the value which was stored was r.eventsRead.Load() which is not correct. Checkpointing is done periodically and generally the value of eventsRead versus eventsApplied will be larger in the PCSM status because some of the events are yet to be applied and PCSM can do a checkpoint with that difference.
But when we perform the recovery, we need to restore eventsRead to be the value of eventsApplied since all the difference, or all the events that we read that did not make to apply when we crashed, are gone and will be re-read when we recover from the last checkpoint.