I've been working more on my experimental test runner and discovered that the sharded test runner isn't always deterministic, at least, not when you invoke it multiple times with parallelism turned on. I'm getting flaky test results.