Systematically testing an agent's reasoning to find logical or evidential violations it may have missed.