I don't think there is any, just full seed collisions. Unless the software pseudo-RNG is so bad that a few million device runs would reveal a lack of uniformity. Perhaps the lesson is that test code random number generators should yield sequential numbers, that are easier to guard against.