One workload replay strategy is to pace the replay to try to bound read-amp to ≤ the read-amp of the reference run (the run from which the workload was collected) at the same event. The manifests in the workload can be used to calculate the read-amplification of the reference run at each step. This will require the replay tool to not entirely skip compaction version edits, and instead apply the edits to a BulkVersionEdit used to represent the state of the reference LSM at the same point. Before performing a Flush or Ingest, the replay tool would wait until its LSM's r-amp is ≤ the read-amp of the reference run at the same point.