We seek to increase user confidence in simulations as they are adapted to meet new requirements. Our approach includes formal representation of uncertainty, lightweight validation, and novel techniques for exploring emergent behavior. Uncertainty representation, using formalisms such as Dempster-Shafer theory, can capture designer insight about uncertainty, enabling formal analysis and improving communication with decision and policy makers. Lightweight validation employs targeted program analysis and automated regression testing to maintain user confidence as adaptations occur. Emergent behavior validation exploits the semi-automatic adaptation capability of COERCE to make exploration of such behavior efficient and productive. We describe our research on these three technologies and their impact on validating dynamically evolving simulations.
Paul F. Reynolds Jr., Michael Spiegel, Xinyu Liu,