Homeostatic Temperature and Primary Emotions as Topological Signatures in Neural Information Flow

We present a pure neural network architecture equipped with a multivariate homeostatic regulation field H(t) ∈ ℝ⁵, and demonstrate that controlled deviations from homeostatic equilibrium produce statistically separable, structurally interpretable topological signatures in the network's attention dynamics. The system contains no symbolic emotion labels, no reward signal, and no external supervision: the emotional analogs emerge entirely from the internal regulatory dynamics of the homeostatic field interacting with attention. Three experimental series are reported. Experiment 1.5 (5 configurations, 50 trials, 40 steps) establishes the baseline result: homeostatic configurations produce significantly different topological signatures across six metrics (p < 10⁻⁸, Kruskal-Wallis). Expansion 2 (8 configurations, 50 trials) demonstrates that FEAR is a continuous field, not a discrete category. Expansion 3 (6 configurations, 30 trials, 80 steps) captures the temporal dynamics of these signatures. This work constitutes Paper I of the HUGO series (Homological Unified Gradient Ontology).Part of the HUGO series: Paper II — Language-Mediated Empathic Coupling (ECHO) Paper III — Persistent Topological Memory.