Include a combination of regular RTS run with a dry run in test/model, with some (presumably) small probability. The combined operation should run RTS in the dry run mode in WT, followed by a regular RTS run, and compare the statistics. The model's implementation would just run regular RTS.