Learning from Vacuously Satisfiable Scenario-Based Specifications

Scenarios and use cases are popular means for supporting requirements elicitation and elaboration. They provide examples of how the system-to-be and its environment can interact. However, such descriptions, when large, are cumbersome to reason about, particularly when they include conditional features such as scenario triggers and use case preconditions. One problem is that they are susceptible to being satisfied vacuously: a system that does not exhibit a scenario's trigger or a use case's precondition, need not provide the behaviour described by the scenario or use case. Vacuously satisfiable scenarios often indicate that the specification is partial and provide an opportunity for further elicitation. They may also indicate conflicting boundary conditions. In this paper we propose a systematic, semi-automated approach for detecting vacuously satisfiable scenarios (using model checking) and computing the scenarios needed to avoid vacuity (using machine learning).