Decision-making has been modeled in great detail based on 2-alternative choice (2AC) tasks; however it remains unclear how these models apply to more naturalistic settings, where choices can have long-term and diverse consequences. In turn, quantitatively modeling more complex decisions poses a challenge, requiring adequate sampling of behavior over a larger state space. To address this problem, we have developed a video game in which subjects can flexibly solve multi-step planning problems that have identifiable optimal solutions. In this game, subjects must plan a path that allows them to collect multiple growing objects at appropriate times in the future. By parametrically varying the number and spatio-temporal arrangement of targets, we densely sample the stimulus space. In parallel, we densely sample behavior by monitoring subjects’ actions and physiology. Each subject develops a highly patterned, yet unique, approach to game play apparent across all measured signals. While some of these signals can be interpreted directly with respect to game play (e.g. button presses), others are indicative of the subjects’ internal state as they solve the task (e.g. pupil size, button pressure). We now seek to combine these two types of measures to understand how subjects efficiently form and update plans.