32 lines
2.1 KiB
Plaintext
32 lines
2.1 KiB
Plaintext
# ─────────────────────────────────────────────────────────────────────────────
|
||
# arms_movement_v2.txt — GATE-V2 fresh-data confirmation of the movement
|
||
# champion, run at the HIGHEST power of the campaign (10 runs/arm vs 5 in the
|
||
# gate-v1 confirmation) on the FROZEN panel tools/ab/panel_movement.txt.
|
||
#
|
||
# Why this file exists (see docs/movement_campaign.md, "Fresh-data confirmation
|
||
# (gate v2)", pre-registered and committed BEFORE these battles):
|
||
# * Gate v1 (commit ff03e81) required BOTH a pooled 95% CI excluding 0 AND the
|
||
# plain cross-opponent sign test at p<0.05. The CI passed; the plain sign
|
||
# test failed at 10/13 decisive, p=0.0923. The plain sign test is the
|
||
# WEAKEST of the three cross-opponent tests the analyzer prints (it discards
|
||
# the magnitude of each paired delta) and was underpowered at n=13.
|
||
# * The gate-v1 failure is NOT reinterpreted. Gate v2 is a NEW
|
||
# pre-registration with the sign-flip permutation test as the PRIMARY test,
|
||
# confirmed on FRESH, independent data.
|
||
#
|
||
# PRIMARY TEST (pre-committed): sign-flip permutation test on per-opponent
|
||
# paired Δwins/run, two-sided p<0.05 AND pooled 95% CI excluding 0 AND a
|
||
# positive point estimate. Ship the default flip ONLY if it passes.
|
||
#
|
||
# REFERENCE is `tfil` — every delta below is (arm − tfil), opponent by opponent.
|
||
#
|
||
# Format: name | ENV=value ENV=value | label
|
||
# ─────────────────────────────────────────────────────────────────────────────
|
||
|
||
# 1. THE CHAMPION under gate-v2 confirmation — current strafe defaults.
|
||
strafe | TR_MOVEMENT=strafe | champion — current strafe defaults (candidate to ship)
|
||
|
||
# 2. THE ARM TO BEAT — the currently shipped default, explicitly selected.
|
||
# (The explicit `TR_MOVEMENT=tfil` also proves the override path stays
|
||
# reachable after any default flip.)
|
||
tfil | TR_MOVEMENT=tfil | shipped baseline — explicit tfil override |