Finicky Sawfishround 3

Hypothesis: The round-2 Elo drop (1518 to 1488) came mainly from the drill. The attacker_stall term rewarded nothing useful and pushed the attacker toward jittery, slow play: mean_speed fell to 42.7 against a league average of 59, with 5.9 direction changes per second. Round 1 had no drill and the same hparams and opponents, and it gained Elo while training stably. Removing the drill and training on outcome only for all 48 iterations should let PPO recover full-speed straight pursuit, since quick tags in the samples come from fast, straight chases. Hparams and opponents stay unchanged so that this round isolates the drill's effect.

Expected signal: Attacker mean_speed rises from 42.7 toward 50-60, and direction_changes_per_sec drops from 5.9 toward 3-4. Attacker win rate against Brave Pronghorn rises from 15% toward 30% and against Classy Cat from 5% toward 15%. Defender results stay within about 5 points; vs Classy Cat, Flexible Caiman and bots-greedy they may improve slightly. Elo +10 to +40. In training, episode_len_mean falls back toward 247 and attacker_win_rate holds above 0.5. If attacker speed does not recover, the jitter is intrinsic to the policy rather than the drill, and round 4 should test a higher entropy or snapshot change.

Diversity: No charter is set. The club stays a plain outcome-driven PPO learner this round, with no shaping. That keeps it distinct from clubs leaning on drills, and it gives a clean baseline before any targeted defender-evasion practice later.