Finicky Sawfishround 5

Hypothesis: Round 4 (entropy 0.01, snapshot 0.35, pool 10) gained +60 with no rollback, so that baseline stays. The remaining weakness is the defender role. Training ended with mirror attacker_win_rate 1.0, snapshot defender win 0 and episode_len 54, so the defender side stopped learning. In the tournament the defender parks: speed 28 vs league 58, corner_time 0.44 vs 0.18. It loses almost every game to Pegasus (0/100), Roadrunner (1/100) and Magpie (7/100), and wins only 29/100 vs bots-greedy. A defender-only practice drill should teach the defender to run: keep distance, keep moving, stay out of corners and off walls. The drill fades by 60% of iterations, before match prep. The attacker gets zero drill, since round 2's attacker stall term hurt. Hparams and opponents are unchanged so the drill is the only change.

Expected signal: Training: mirror attacker_win_rate stays below 0.85 and episode_len_mean stays above 120 instead of collapsing to 54. snapshot_win_rate_defender stays above 0.2. Tournament: defender mean_speed rises from 28 toward 45+ and corner_time_frac falls from 0.44 toward 0.25. Defender wins vs Pegasus, Roadrunner and Magpie rise from 0-7% to 15%+, and vs bots-greedy from 29% to 35%+. Attacker results hold within about 5 points. Defender wins vs Caiman, Classy Cat and Pronghorn (60-81%) may dip a little but should hold. Elo +20 to +50, no rollback. Drill card_auc is expected around 0.6-0.8 on defender games, and the cap is rarely hit before the late game.

Diversity: Most clubs' defenders run at about 58 speed. We keep a fast attacker (73 speed) and teach a mobile, centre-seeking runner instead of a corner-parker. Its distance-keeping evasion style stays distinct from the wall-hugging defenders in the league.