PHILIA-EcoSensory Swarm v50: The Linear Geometry of PHI_NATURAL — z = 0.42 − 0.22·w Across Three Architectures

We present PHILIA-EcoSensory Swarm v50, a comprehensive longitudinal self-audit of 400 runs across eight controlled experiments (Tier A, Rec1–Rec7), establishing the causal mechanism underlying PHI_NATURAL z-axis drift observed across PHILIA versions v32–v49. Through a complete w-clamp ladder (Rec7: L0/L1/L2 at w=0/0.5/1.0), we confirm the linear z-attractor equation: z(w) = 0.4234 − 0.2159·w (R² = 0.991) This relationship holds across three independently developed architectures (v32 single Phi_center, v34 output-blend, v49 3-Gearbox input-blend). Cross-architecture invariance is most striking at w=1.0 (Taxi only): v34 z=0.2019 vs v50 z=0.2016 (Δ=0.0003), despite different hardware (Intel N100 vs Ryzen 9800X3D) and seed sets (N=5 vs N=20). Key findings:(1) PHI_NATURAL = [0.34, 0.19, 0.42] is the w=0 limit of a continuous attractor family, not an architecture-specific eigenstate.(2) >85% of z drift is caused by direct Taxi (NYC Taxi) input entropy injection (I_eff → S_history → H_i → z = 1 − H_i/H_max). z_KL and Neuron type contribute negligibly (<1%, 0%).(3) Two causality regimes identified: v34 temporal corr(w,z) and v50 stationary mean(z|w_clamped) are complementary — Being (geometric law) vs Becoming (dynamic response).(4) v49 Neuron class modular cleanness empirically verified: PhiliaV35Neuron with z_KL/PPG OFF reduces algebraically to PhiliaV32Neuron_Simple. All results from local machine execution (AMD Ryzen 9 9800X3D + RTX 4080 SUPER 16GB). No AI-generated or proxy values used. 실측이 생명. 증명이 아닌 기술.