Results

• Performance in leader-follower problem with a worst-case payoff matrix---the leader maximizes his WL reward when his is placed in separate subworld from followers.
• The top curve represents perfect constraint alignment in the initial subworld-assignment (i.e., leader and followers in the same subworld), the bottom minimal constraint-alignment (i.e., leader and followers in different subworlds), and the middle random subworld assignments.
• There is no macrolearning.

• With macro learning at 500 weeks on the middle curve (random assignments).
• The random subworld-assignment jumps when macrolearning happens.