[rscyqzkb] Paradoxical Go handicaps

Given a set of games between people with ratings (chess) or rankings (Go 囲碁 ), find large subsets of players for which all the games played among the subset do not conform well to the predicted outcome based on the ratings or rankings.  If such subsets exist, it is evidence that a single parameter, abstractly "strength", is not enough to predict outcomes well.

One specific scenario has the subset further partitioned into 3, and each partition tends to beat the next in a cyclically, like rock paper scissors.

Go 囲碁 rankings are potentially useful because they are calibrated to handicaps, allowing players of different strength to play an even game.  Though one could imagine confounding factor that some people may be able to leverage a handicap better than others, or play worse against a handicap than others.  For a 1-stone handicap, i.e., always play black, would be especially helpful for a player good at maintaining initiative and turning initiative into wins.