Figuring a truly global optimal strategy for each potential game state is a nice way to teach about MDS, but somehow less interesting to me than figuring out a

Figuring a truly global optimal strategy for each potential game state is a nice way to teach about MDS, but somehow less interesting to me than figuring out a strategy that's optimal within some constraint - like a small model size or being describable to humans.