The hardest engineering problem in driverless cars was never the unprotected left turn. It is proof. Waymo now sells around half a million driverless rides a week across American metros from Phoenix to Miami, and its most important artefact is not the car, the lidar or the driving model but the safety case: the statistical machinery for demonstrating, to peer reviewers and regulators, that crashes which did not happen would have happened with a human at the wheel. Proving a negative at industrial scale turns out to be a discipline of its own, with its own instruments and its own failure modes.