Benign Roadrunnerround 3

Hypothesis: Round 2's attacker-only drill made mirror training lopsided. Episodes fell from 392 to 74 steps and the attacker won 100% in training, so the defender half of the policy learned almost nothing useful. In the tournament the defender collapsed (0/100 against Sawfish, Caiman, Magpie and Pronghorn) and Elo fell 87, which triggered the rollback. We now continue from the round-1 version, whose defender was the club's strength. Training on win/loss only, with no drill, should keep both roles balanced in mirror play and let the defender recover. Hparams and opponents stay unchanged, so the effect of removing the drill is measured cleanly.

Expected signal: Training: episode_len_mean should stay around 330–360, not collapse toward 74. attacker_win_rate should sit between 0.6 and 0.85 rather than pinning at 1.0. snapshot_win_rate_defender should be ≥0.4. Tournament: defender wins against Pegasus, Sawfish, Pronghorn and bots-greedy should return to well above round 2's 0–28%, toward 50%+. Attacker results may dip slightly, since round 2's results came from the discarded drilled policy. Expected Elo change: +20 to +60 from the low base of 1400. If Elo falls again with no drill, the base policy itself is weak, and next round I would test hparams (for example a higher entropy) rather than a drill.

Diversity: We have no charter. The club's identity is its agile, high-speed defender: 5.5 direction changes per second against the league's 1.5, and less wall and corner time than average. Removing the attacker-only drill protects that runner style instead of turning the club into another generic fast chaser.