It is known that thermal perception influences occupants’ adaptive actions in buildings which in turn influence their thermal satisfaction. During recent years there has thus been a growing interest in the development of models which predict occupants thermal comfort, accounting for subtleties such as the transient nature of the environment inside free-running buildings. But all of these empirical models predict a mean neutral temperature, for a given weighted mean of external temperature for an entire population, irrespective of the specific adaptive actions exercised. As such they do not deepen our understanding of the underlying mechanisms and they are readily applicable to other contexts which may not share the same features as these from which the models were developed. In this paper we attempt to resolve this lacuna. We first present a model to predict a probability distribution of thermal sensation in free-running buildings. We then introduce a methodology for combining recent advances in predicting occupants’ adaptive behaviour with prediction of the thermal feedback from this latter. We go on to demonstrate the explanatory power of this new methodology and how thermal comfort and behaviour are linked by the concept of adaptive inertia. We also address ways of handling individuals’ diversity with this new framework as well as how it may be applied to other comfort (eg. visual) domains. We believe that this new approach considerably deepens our understanding of the general subject of adaptive comfort whilst also offering a solid basis for application to buildings not in the calibration dataset.