This paper reports the first fully executed local 120-window verification run of the Barker three-variable coronal mass ejection (CME) eruption score. The benchmark now contains 40 eruptive windows, 40 flare-rich confined windows, and 40 quiet/control windows drawn4 across 102 HARP groups. The score decomposes eruptive readiness into a system-scale magnetic coherence term C, a localized AIA 304 Å loading term n, and an overlying-field confinement term H built from a potential-field decay construction above the strong-field polarity inversion line. The benchmark is scored at the window level rather than by pooling autocorrelated hourly rows, because the scientific question is whether entire active-region windows separate cleanly. The primary benchmark in this paper is the window-median comparison of eruptive versus non-eruptive windows, where the non-eruptive pool includes both confined and quiet/control cases. On that test, the baseline 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 SnH = log 1 + n1.5 (H+ 0.05)2 reaches AUC 0.3282 with mean gap−0.3664, whereas the corrected score Scorr = log 1 + Cn1.5 (H+ 0.05)2 reaches AUC 0.5161 and mean gap 0.0173 when C= USFLUX. The resulting deltas are +0.1880 in AUC and +0.3837 in mean gap, with HARP-group bootstrap intervals that remain strictly positive: [0.1152, 0.2658] for the AUC delta and [0.2185, 0.5643] for the mean-gap delta. The original pilot challenger C= TOTUSJH also generalizes positively at scale, reaching AUC 0.5041 and mean gap 0.0115 on the same primary benchmark. TOTPOT performs at nearly the same level (AUC 0.5051, mean gap 0.0063), confirming that the winning broad-batch coherence family is compact rather than brittle. Against the stricter eruptive-versus-confined contrast, the corrected score remains positive. The strongest mean-collapsed result is obtained with C= TOTPOT, which lifts AUC from 0.4431 to 0.5113 and mean gap from−0.0387 to 0.0104. The executed 120-window run therefore does not merely keep the three-variable framework alive; it shows that once the benchmark is evaluated at the intended window level, the corrected score materially outperforms the two-variable baseline across the broad non-eruptive pool and remains positive on the harder confined-only stress test.