I guess it starts to become a bit complicated for people because they will need to have a lookup table etc. Effectively the simulator is doing this an more when it plays with one fixed word because the second word is chosen based on what it learns from that fixed word. So this should do better than what you describe since it can select from a larger set of second words. However, the results show that there is at least a single best combination for two words which does as well as the simulator when used with a single best word and a carefully chosen game-specific second word. So I think your strategy cannot do better that the simulator in single-word mode and wont be materially better than the simulator when it uses its best two words. However, and this is a bit ‘but’, the simulator is a close to perfect player and this is not reasonable. So it may well be the case that what you describe worlds well in practice.