PHILIA v11–v13: Individuality Collapse Mechanism in Recursive Homeostatic Systems

📋 Zenodo 설명란 (Description) PHILIA v11–v13 presents a systematic three-version elimination study identifying the exact mechanism of individuality collapse in recursive homeostatic multi-agent systems. Building on PHILIA v11's baseline finding — that a goal-adaptive system of N = 50 agents achieves robust homeostasis but loses all agent-level diversity within ~5,000 of 60,000 timesteps — this paper conducts controlled single-variable experiments to isolate the root cause: v12 tests whether input homogeneity causes collapse by splitting data across agents (tile → split). Result: collapse unchanged. Input diversity is not the cause. v13 tests whether global mean-field coupling causes collapse by setting λ = 0. Result: collapse unchanged. Global coupling is not the cause. S-variance diagnostic measures structural state S directly across all 50 agents. Result: S itself converges to a single value within ~2,000 steps — preceding Goal Variance collapse — confirming that the contractive structural update dynamics are the upstream cause. The complete causal chain is established: shared normalization range → similar Φ values → S converges to single attractor → identical Reward functions → Goal synchronization regardless of input diversity or coupling structure. Core finding: "Individuality is determined not by input diversity or inter-agent coupling, but by the single-attractor nature of the structural dynamics." All numerical results are ground-truth outputs executed directly on the local machine of the lead researcher (GritMan_D.S, Trinity AI). Part of the ongoing PHILIA Engine research series.