Non-functional properties, such as reliability and performance, should be assessed as early as possible in a system's life cycle for cost effectiveness reasons. Hence, several software architecture-based reliability assessment techniques have been proposed. These techniques quantitatively analyze a system's or a component's behavior, which is typically represented using probabilistic generative automata. However, we demonstrate that generative automata do not appropriately capture information available in an operational profile. Furthermore, we overview other existing probabilistic automata formalisms; we identify their features and shortcomings when capturing an operational profile. As a way to circumvent the identified deficiencies, we introduce probabilistic component interface protocols, a new probabilistic automata formalism that supports intuitive and direct mapping of an operational profile. Finally, we discuss how to derive the analysis-oriented generative models from the probabilistic component interface protocols.