It's interesting to see the averages from different decision trees and that the optimal hard mode is only 3.51 -- and that easy/hard mode are so close. It makes me wonder how much of my stats are pure luck. I'm currently sitting at exactly 3.5 after 48 hard modes (0/2/26/14/6/0) starting with SLATE, but while I feel pretty confident playing the game, I definitely have not used the "optimal strategy" every time, if I even knew that strategy in the first place.

Given how close hard mode and easy mode are, and that even when calculating averages, 4, 5 & 6 are still considered "wins", I strongly suspect changing the parameters could show hard mode to be "easier" for a "<4 or bust" goal.

> I definitely have not used the "optimal strategy" every time, if I even knew that strategy in the first place.

It's also worth noting that these models (presumably) assume an equal chance for every word in the provided dictionary, whereas - assuming the answers are "common" words - a human will likely be strongly biased toward the subset of possible answers and therefore can have much higher probabilities of success in practice than the theoretically-optimal strategies.

TALES is almost as good as a first word, and isn't obtuse.

"while we don't know that "SALET" won't someday be an answer"

Yes, we do ... all the answers and the day they will occur are known.

Was phrasing the statement from the perspective of a model "player", but I guess knowing whether "SALET" is in the answer list is not quite the same as knowing where each word is in the list, so you could semi-cheat and use the answer list as your definition of "common", without gaining too much knowledge.