At the moment, models are evaluated on their ability to fit the existing data, rather than forecast (with an implicit assumption that there is a simple relationship between them). Having single/multi-step machinery (which is framework-specific) would allow us to address this.