With trillions upon trillions of possibilities, how could you identify a set of test cases for an autonomous system that is both manageable and adequate?