BKT and other classical student models are designed for binary environments where actions are either correct or incorrect. These models face limitations in open-ended and data-driven environments where actions may be correct but non-ideal or where there may even be degrees of error. In this paper we present BKT-SR and RKT-SR: extensions of the existing BKT model that distinguish knowing "how" to apply a skill from knowing "when." We compare their relative performance to that of classical BKT and PFA on data collected from Deep Thought, an open-ended propositional logic tutor. We develop basic performance curves for student outcomes to help us visually compare models predictions to data. We also introduce a new approach for finding a probability distribution of actions in ranked, multiple option environments with RKT and RKT-SR. Our results show that knowing when to use skills is more important than how in these open-ended contexts. In fact, including the "how" components may hurt performance if implemented naively. Furthermore we show that ranked models outperform binary-based models even under restrictive assumptions. [For the full proceedings, see ED596512.]