We formally specify the interpretation stage in a dual state space human-computer interaction cycle. This is done by extending / reorganising our previous cognitive architecture. In particular, we focus on shape related aspects of the interpretation process associated with device input prompts. A cash-point example illustrates our approach. Using the SAL model checking environment, we show how the extended cognitive architecture facilitates detection of prompt-shape induced human error. Key words: human error, cognitive architecture, model checking, SAL.