Quite a bit of work and thought gone into this.
Perhaps you could normalize your WAR statistic somehow by normalizing using number of battles, and earliness or lateness of first loss in battle. This would give Alexander an advantage over Hannibal. May not change the ranking but may provide a more accurate picture. It is not impossible also that a quadratic function may fit the data better than a linear function. But then if you are a General you can’t refuse to fight if you are attacked on basis of you have been doing that a long time. A liner function may still work better, but my research mind continues to wonder how well a quadratic function would perform in comparison with your linear model. Not saying you should definitely check it out, just some thoughts crossing my mind.
Interesting stuff. Was skeptical when I started reading, wondered whether you really had pulled this off. You won me over because now my mind is all over the place thinking how the model could be fine tuned to perform even better.