Some flake swarming tasks time out, however still have usable output. Consider capturing this output to determine a pass rate. If the test is obviously flaky from the salvaged output there's no need to rerun and the analysis can actually continue.