We should write an FSM workload that verifies read repeatability of transactions. This workload can presumably have each thread be in either a Read or Update state, where the Read state executes multiples reads sequentially, expecting to see the same result set for each read. The Update state could update some random subset of documents in a collection. This test would be good at verifying repeatability under higher concurrency and load than our targeted tests. Eventually we may also add a repeatability test that runs against all our existing FSM workloads, but in lieu of that, this could be a valuable targeted workload to exercise a key property of transactions under snapshot isolation.