Fair enough. My thinking was that you'd use probabilistic methods and run many trials to see where the failures are, but I guess that's likely to fail if the re

Fair enough. My thinking was that you'd use probabilistic methods and run many trials to see where the failures are, but I guess that's likely to fail if the real-world patterns are too cross-correlated, which they undoubtedly are (rush hour, etc.).