In "How Task Features Impact Evidence from Assessments Embedded in Simulations and Games," Almond, Kim, Velasquez, and Shute have prepared a thought-provoking piece contrasting the roles of task model variables in a traditional assessment of mathematics word problems to their roles in "Newton's Playground," a game designed to assess conceptual understanding of mechanics, creativity, and persistence. The article addresses a number of themes; this commentary focuses on variant generation and task difficulty and the potential role of automatic generation therein. Although variant generation is considered primarily with respect to mathematics word problems, it has the potential to be applied to games with embedded assessments as well. Some ideas for how automatic generation might support difficulty modeling and development for games with embedded assessments are proposed. Final comments concern the use of construct maps to characterize proficiency.