 We developed a novel methodology to assess the predictive power of non-identifiable computational models. This methodology involves measuring a single variable in response to a carefully designed stimulus protocol, which reduces the dimensionality of the parameter space and enables accurate prediction of the measured variable's trajectory under different stimuli. Furthermore, this methodology can also be used to predict how a given trajectory would evolve if any of the model parameters were changed. Finally, successive measurements of other variables can further reduce the dimensionality of the parameter space and allow for more accurate predictions. This article was authored by Fredrik Grabowski, Paol Naleks-Jwecki, and Tomasz Lipniakki.