We focus on non-functional requirements for applications offered by service integrators; i.e., software that delivers service by composing services, independently developed, managed, and evolved by other service providers. In particular, we focus on requirements expressed in a probabilistic manner, such as reliability or performance. We illustrate a unified approach—a method and its support tools—which facilitates reasoning about requirements satisfaction as the system evolves dynamically. The approach relies on run-time monitoring and uses the data collected by the probes to detect if the behavior of the open environment in which the application is situated, such as usage profile or the external services currently bound to the application, deviates from the initially stated assumptions and whether this can lead to a failure of the application. This is achieved by keeping a model of the application alive at run time, automatically updating its parameters to re?ect changes in the external world, and using the model’s predictive capabilities to anticipate future failures, thus enabling suitable recovery plans.