This paper discusses the central problems with creating valid computational social science simulations, and then suggests answers to these problems that involve co-evolution, autonomy, interpretation, and data processing under uncertainty. We also present the use of these techniques in the Nexus cognitive agent simulation, used at the US Department of Defense (DoD) in multiple major analyses of Irregular Warfare. We introduce the technique of data absorption, which leverages co-evolutionary pressure to reproduce the same dynamic structures that caused observable real word data in the simulation through the motivations of the agents.This technique gives a causal explanation for the data, and sets the stage for testing the effects of interventions never seen by the system on the system. By mimicking the state of equilibrium reached by the natural system, data absorption closes the gap between theory-centric simulations and data centric simulations.