The "evidence-centered design" (ECD) framework is a powerful tool that supports careful and critical thinking about the identification and accumulation of evidence in assessment contexts. In this paper, we demonstrate how the ECD framework provides critical support for designing simulation studies to investigate statistical methods within an ill-defined methodological domain like games-based assessment. We discuss the design and selected findings of a complex simulation study to investigate the utility of statistics derived from a non -parametric method called "epistemic network analysis"; this method is used in practice for the analysis of real data from a suite of digital learning environments called "epistemic games." We present findings that show that one marginal ENA statistic, the "weighted density" statistic from social network analysis, has some ability to detect differences among a variety of simulated learners when they play different types of simulated games. Detection strength ranged from weak to strong, depending jointly on game design characteristics and the types of learners being compared. Our work illustrates the complex challenges of how best to describe, justify, and evaluate design decisions for simulation studies in the context of games-based assessment.