On formalising interactive number entry on infusion pumps

We define the predictability of a user interface as the property that an idealised user can predict with sufficient certainty the effect of any action in a given state in a system, where state information is inferred from the perceptible output of the system. In our definition, the user is not required to have full knowledge of a history of actions from an initial state to the current state. Typically such definitions rely on cognitive and knowledge assumptions; in this paper we explore the notion in the situation where the user is an idealised expert and understands perfectly how the device works. In this situation predictability concerns whether the user can tell what state the device is in and accurately predict the consequences of an action from that state simply by looking at the device; normal human users can certainly do no better. We give a formal definition of predictability in higher order logic and explore how real systems can be verified against the property. We specify two real number entry interfaces in the healthcare domain (drug infusion pumps) as case studies of predictable and unpredictable user interfaces. We analyse the specifications with respect to our formal definition of predictability and thus show how to make unpredictable systems predictable.