[NathanOsullivan]

I agree, the answer list is clearly curated to avoid unusual words.

This leaves a fuzzy middle ground for human players: there are accepted guess words like SOARE that can be good as a first guess, but is almost certainly not going to be the answer. Conversely, FAVOR is not a good first guess; but of the two words human players know which is more likely to be the answer.

For computer implementations that want to avoid "cheating" (not use the answer list), there would still seem to be room for evaluating how likely a given guess is to be in the curated list.

That could be done by examining a corpus for frequency of each guess word, which if done right should give the same insight we have as humans.