Predictability of Enforcement

摘要

The current theory of runtime enforcement is based on two properties for evaluating an enforcement mechanism: soundness and transparency. Soundness defines that the output is always good ("no bad traces slip out") and transparency defines that good input is not changed ("no surprises on good traces"). However, in practical applications it is also important to specify how bad traces are fixed so that the system exhibits a reasonable behavior. We propose a new notion of predictability which can be defined in the same spirit of continuity in real-functions calculus. It defines that there are "no surprises on bad input". We discuss this idea based on the feedback of an industrial case study on e-Health.