I worked down the hall from Ken Arnold (he was one of the later additions to the project, mentioned about half way down the page) at HP/Apollo around 1990; aside from being an inspirational Senior Troublemaker, he kept an eye on Rogomatic - an advanced rogue-playing "bot" - and would make minor tweaks to Rogue itself that didn't change things for human players, but were adversarial for the bot...
That's fascinating! Do you recall any examples of such tweaks? To my mind, making the decision more complicated for all players, human and bot alike, would probably have been justified. But I suppose for a 1980s computer scientist targeting only the bot was an extra-interesting puzzle.
