Across mixed-order simulations, total accuracy changes little by section type, but the first three questions after each section transition show a consistent two-error spike. What should the student add?