SirStone
0e7e124c7f
Pooled verdict on the pre-registered rule is NOT DISTINGUISHABLE: the primary cross-opponent sign test on round wins is 11/14, p=0.05737 for 'arrive', over the 0.05 line. Block 1 alone passed (p=0.01294); block 2 alone did not (p=0.0654). Every delta is positive in both blocks for all three arms and damage/run is UP on all three, so this reads as an under-powered null at the MDE boundary (MDE 0.31 wins/run, observed 0.28) - but the verdict layer is not reinterpreted, exactly as gate v1 was not. The MECHANISM is established cleanly and replicates in both blocks: incoming hit rate 18.07% -> 14.92%, sign-flip p=0.0013, CI [-5.29, -1.53] pp, damage taken -24.57/run. That is precisely the pathology the owner watched. Recommended for his own .env: TR_TFIL_COMMIT_ARRIVAL=1 and TR_TFIL_NOREV_SPEED=4; leave TR_TFIL_COMMIT_MARGIN at 0 (weakest arm in both blocks). Shipped default untouched: TR_MOVEMENT=strafe, all three new knobs default off.