This paper describes one aspect of recent work in the RAND Strategy Assessment Center (RSAC), the challenge of finding ways to incorporate military realism in analytically oriented automated war games. Meeting the challenge has required developing new concepts and techniques that show great promise. The philosophy behind them has even greater applicability. One such concept is that of analytic war plans, logic structures attempting to capture the many high-level decision points that would be faced by the United States and Soviet Union in conflict. The analytic war plans are abstract rule-based generalizations of decision trees. The other concept treated in this paper defines an approach to combat modeling that sacrifices analytic elegance for pragmatism as part of a philosophy requiring the RSAC to treat numerous special phenomena of war that can have important implications even at the strategy level, but which have traditionally been ignored or poorly treated in analytic models.