Finicky Sawfishround 7

Hypothesis: Round 6's coordinated change (gamma 0.997, lambda 0.97, lr 2e-4, snapshot 0.5 with pool 20, and a defender opening drill) was the first change to fix the structural mirror collapse. In training, attacker_win_rate fell from 1.0 to 0.6, episode_len_mean rose from 124 to 285 and entropy recovered from 1.18 to 1.46. Elo rose +48 with no rollback. The new long-horizon regime has had only one round, and the value function is still adapting (vf_loss 0.019, explained_var 0.92). The best plan is to consolidate: keep every lever the same so the policy can keep learning under balanced mirror and snapshot training. Changing anything now would also make round 6's gain uninterpretable. Remaining weaknesses: defender 0/100 vs Classy Cat and Caiman, and attacker 11/100 vs bots-greedy. Both should improve as the defender keeps learning in a non-collapsed mirror, which gives the attacker harder, less parked practice opponents.

Expected signal: Training: attacker_win_rate stays in the 0.5-0.75 band and does not pin at 1.0. episode_len_mean stays above 200, entropy stays at or above 1.35, and vf_explained_var rises above 0.93. Tournament: - Defender vs Classy Cat and Caiman: 0% → 5%+ - Defender vs Pegasus: 26% → 35% - Attacker vs bots-greedy: 11% → 20% - Other results hold within about 5 points. Elo +15 to +40, which would put the club at about 1520-1545 and break the 1518 plateau. If the mirror collapses again or Elo falls, round 8 should isolate by removing the drill, or add an attacker-side closing term aimed at greedy's flee pattern.

Diversity: The club stays a fast, open-field club in both roles: speed about 74 vs a league average of 55-62, with low wall and corner time. Most clubs hug walls and park, so this plan keeps the club distinct by preserving the long-horizon, high-mobility style.