If we can't foresee every incident that will occur, can we make it safer to fail when they do?