Flexible Caimanround 2

Hypothesis: Self-play collapsed into attacker dominance (mirror attacker win rate 0.9) with defenders freezing in corners, so our attacker never learned to finish a moving or well-placed defender (2/100 vs Sawfish and Pegasus) and our defender stalls. A distance drill helps both roles: attackers are rewarded for closing distance, defenders for opening it, plus a small mobility bonus for defenders. The distance term telescopes over a game, so its total is bounded. With slightly more entropy and more snapshot games, the defender should toughen up and give the attacker a harder practice partner.

Expected signal: Mirror attacker_win_rate falls from 0.9 toward 0.6-0.7 and episode_len_mean rises to about 280+ during training. In the tournament, attacker wins against Sawfish, Pegasus and Classy Cat rise from about 2-11 to 15+ per 100, and defender wins against bots-greedy rise from 19 to about 30. Overall Elo is recovered by +30 to +60. card_auc should come out above 0.6.

Diversity: No charter. The club leans toward a fast, low-direction-change style: we push mobile pursuit and evasion rather than the league's common corner-camping defence.