We consider the problem of modelling and verifying the behaviour of systems characterised by a close interaction of a program with the environment. We propose to model the program-environment interplay in terms of the probabilistic modifications they induce on a set of application-relevant data, called data space. The behaviour of a system is thus identified with the probabilistic evolution of the initial data space. Then, we introduce a metric, called evolution metric, measuring the differences in the evolution sequences of systems and that can be used for system verification as it allows for expressing how well the program is fulfilling its tasks. We use the metric to express the properties of adaptability and reliability of a program, which allow us to identify potential critical issues of it w.r.t. changes in the initial environmental conditions. We also propose an algorithm, based on statistical inference, for the evaluation of the evolution metric.