Counterexample reduction in the model-based testing currently removes unnecessary workload operations that do not contribute to reproducing a failed test run, with the goal of producing the minimal reproducer. The current algorithm only works at the transaction level, which results in a reproducer with just the necessary transactions and other "high-level" operations such as setting the stable timestamp.

Reduce the counterexample further by 1) eliminating tables and the corresponding operations that do not contribute to reproducing the failure, and 2) also removing as many operations from within the transactions as possible.