Experiment 02 · fixed 24×24 world

Team-size sweep @ 24²

Holds the 24×24 world fixed and grows the team from 4 to 12 agents, comparing the homogeneous shared policy against the learned skill-selector variants — the question being whether extra agents divide the labor or simply flood the same map with redundant coverage.

Key finding The homogeneous shared policy scales up with the team and leads at 9–12 agents, while the congestion-priced skill-selector variants fall off as the team grows. Total coverage rises with more agents, but the gain is sub-linear — consistent with added agents increasing redundancy rather than cleanly partitioning the map.
t = 0