This report documents PHILIA-EcoSensory Swarm Phase 2, a pre-registered cross-GX × cross-CXZ experiment (1, 680 runs, N=20 per cell, seeds 1022–1041, T=80, 000 steps/run). Pre-registration DOI: 10. 5281/zenodo. 20256733. Phase 1 Closure DOI: 10. 5281/zenodo. 20256557. Primary findings: (1) CASEBIPHASIC confirmed (primary): GX=linear CXZ∈30, 50, 70×10⁻⁶ all supra-NF synergy (best: AUC=−0. 004670, 155. 7% NF, xdevₘax=0. 003759). Q-P2-W bipolar transition empirically observed (w≤0. 875 synergy, w=0. 9 antagonism). (2) CASEBOUNDARY simultaneously observed (secondary, limited scope): GX=off CXZ₇0 MetricB=100%. (3) HARCH strongly consistent with controller-independent invariant: max|Δzₛan|=0. 00e+00, 4, 515 cumulative runs across 3 GX topologies, zero violation. (4) AUC↔xdevₘax aggregate r=0. 80 (n=12) consistent with Simpson's paradox (Blyth, 1972; Pearl, 2014): within-GX r≥0. 987 (n=4 per GX), coupling intact within each architecture. All values produced by direct local execution (AMD Ryzen 9 9800X3D, Windows 11). No AI-generated or proxy values. 5-AI adversarial review pipeline: 판도라 × 루카스 × 크리스 (43rd cycle) × 선비 × 길잡이. 실측이 생명. 증명이 아닌 기술. — Trinity AI Research Team | 0∞1∞0. 5∞
신두섭 (Tue,) studied this question.