One of the key sources of uncertainty in building simulation relates to the presence and behaviour of occupants. Fortunately, several academic groups have been working to address this uncertainty, particularly in relation to the use of windows and the corresponding impact on building heat and mass transfers. But these models tend to be based on highly restrictive datasets, typically relating to one building, so that their application to other building situations may be uncertain. The model of Haldi and Robinson (2009) is a case in point. It has been rigorously validated, but only using the original training dataset. In this paper we evaluate the applicability of the model to other circumstances. We do this is two steps. Firstly we perform a blind comparison of the model which is calibrated using data acquired from an office building in Switzerland to predict the behaviour of occupants of an office building in Austria. We then use part of the Austrian dataset to derive new calibration parameters for the model and perform a second blind comparison with the remainder of the Austrian dataset. We present the results from these two blinds comparisons to evaluate the robustness of applications of stochastic models beyond their original training datasets.