To be fully effective, combat simulation must include an intelligently interactive enemy... one that can be calibrated. But human operated combat simulations are uncalibratable, for we learn during the engagement, there's no average enemy, and we cannot replicate their culture/personality. Rule-based combat simulations (expert systems) are not interactive. They do not take advantage of unexpected mistakes, learn, innovate, and reflect the changing mission/situation. And it is presumed that the enemy does not have a copy of the rules, that the available experts are good enough, that they know why they did what they did, that their combat experience provides a sufficient sample and that we know how to combine the rules offered by differing experts. Indeed, expert systems become increasingly complex, costly to develop, and brittle. They have face validity but may be misleading. In contrast, intelligently interactive combat simulation is purpose- driven. Each player is given a well-defined mission, reference to the available weapons/platforms, their dynamics, and the sensed environment. Optimal tactics are discovered online and in real-time by simulating phenotypic evolution in fast time. The initial behaviors are generated randomly or include hints. The process then learns without instruction. The Valuated State Space Approach provides a convenient way to represent any purpose/mission. Evolutionary programming searches the domain of possible tactics in a highly efficient manner. Coupled together, these provide a basis for cruise missile mission planning, and for driving tank warfare simulation. This approach is now being explored to benefit Air Force simulations by a shell that can enhance the original simulation.