This paper presents the foundations of a computational science of game design—a model of abstraction: the outcome of a situated design process that conceptualizes something in terms of something else for a specific purpose. The outcome of abstraction is modeled via an abstraction scheme, a step toward modeling the process of game design abstraction from a human-centered perspective. A scheme’s purpose is couched relative to a scheme’s properties, also defined herein; collectively, the properties represent the space of tradeoffs designers must navigate for engineering virtual worlds. This model’s analytical traction is evidenced by applying it to the design of pathfinding, a core behavior in artificial intelligence for games. More broadly, it is a foundation for shared progress because it affords directly comparing particular abstractions of concepts and phenomena across the gamut of research on intelligent systems in entertainment.