ANIMA: Longitudinal Cognitive Advantage Through Homeostatic Field Persistence — Dual-Mode Ablation Study with Behavioral Trait Analysis

We present ANIMA (Autonomous Neurodynamic Integration for Modular Agents), a protocol for longitudinal cognitive evaluation of agents equipped with the HUGO homeostatic field. Two experiments totaling 296 sessions on Llama 3.1 8B:- ANIMA-02: Dual-mode ablation (200 sessions) — 4 arms × 5 seeds × 2 modes (standard γ=0.97, persistent γ=0.995)- ANIMA-03: Gamma sweep (96 sessions) — 8 γ values × 3 seeds × 2 passes × 2 sessions Key findings:- γ is a personality parameter, not a hyperparameter — different values produce qualitatively different agents- Peak learning at γ=0.990 (PSI=+0.359, 36% improvement between passes), not at extremes- 7 empirical cognitive archetypes confirmed: Cold Reactor, Warm Improviser, Hot Improviser, Adaptive Coordinator, Deep Learner, Guardian, Contemplative- 14 observable behavioral traits cataloged from session telemetry- Heterogeneous agent teams with varied γ values outperform homogeneous teams Related software: https://github.com/aprimora-ai/hugo-framework-anima (DOI: 10.5281/zenodo.19111951)