Scientific theories aim to provide a framework for understanding several phenomena by discretizing the universe and its nature into well-defined concepts, and formulating their interrelation. Unfortunately, this deterministic approach towards understanding is not always justified, and often conflicts with the uncertainty around stochastic processes and chaotical behaviour. Consequently, quantum mechanics and general relativity have superseded the Euclidian dimensions of space(-time), and the popularity of probabilistic theory has considerably increased. Prediction models have become operating theories of the 21st century, and are increasingly based on large amounts of observations. Unfortunately, numerous prediction models do not generalize well and perform more poorly than anticipated when applied in real life settings.
I firmly believe that evidence synthesis is needed to improve prediction model performance and to anticipate for potential deficiencies that may arise when implementing a model in particular settings. Furthermore, external validation is crucial to identify the extent of a model's generalizability and expose its boundaries of applicability.