Adversarially assess whether a verified bug primitive can be exercised by a realistic attacker in production and produce the claimed security impact with measurable reliability.
