1: Default is to play small (conserve points). As long as the effect of a round is not “the lowest reputation player is affected after a pass”, the tree almost always chooses your smallest remaining card, across most reputation and target-average settings. This is the dominant behavior in the largest branch of the tree. 2: If your reputation is negative and the round target average is low, spend big to stabilize. When your reputation is negative and the target average per player is 4.5 or lower, the tree often switches to your largest remaining card (especially if your remaining cards are all fairly high already). 3: When a fail is heavily punished, play middle cards more often. If the effect of a round is “the lowest reputation player is affected after a pass” and the reputation change after a fail is very negative (around -1.5 or worse), the tree prefers a middle card rather than the smallest. 4: In high-average rounds, prefer middle cards. When the target average per player is above ~6.5, the tree leans strongly toward middle card choices (with a few “still play small” exceptions when other conditions make passing easy).
States evaluated: 18759
Simplified rules match the full tree: 0.775
This is agreement with the full decision tree tiers, not tournament strength.
The simplified model outputs a tier:
This is meant to be a readable mental model. It will lose strength compared to the neural policy.