SOCIETAS: Govern a Society With a Mind of Its Own

SOCIETAS is a real-time, explainable governance simulator: a human types a policy and a population of 80 to 1,000 autonomous agents on a toroidal 30x30 world answers back, not with a spreadsheet but with a foodstaple vendor's panic, an unemployed parent's theft, a gang's recruitment drive, a doctor's burnout, a marriage, a headline. Beneath the drama sits a deterministic-then-LLM contract: Layer 1 is the Cartesian body, pure math with no free will, running a 17-step tick from policy through aging, needs, welfare, emotion, gangs, action, movement, riot, death and birth, closed by SHA-256 hashing; Layer 2 is the Cartesian mind, three Gemma 4 models reasoning and narrating what the math already decided. Each agent carries 13 Maslow needs, 8 Beta-distributed traits, and a five-state emotion machine driven by a Freudian unlust equation weighting food, water, safety, social, and financial deprivation, gated by morality and watched for a Thanatos death-drive signal. Adlerian striving scores social standing, driving ambition or gang leadership; Marxian conditions sort agents into four wealth classes whose deprivation breeds protest, riot, and crime. Governance ripples through progressive tax, welfare, and labor and property markets across 24 actions, 15 professions, and 7 death causes. The stack runs on one AMD Instinct MI300X, 192GB HBM3, holding a 31B dense model, a 26B A4B mixture-of-experts moral-dilemma detector, and an E2B brain model live together, no swapping, sub-2s completions, zero failures across 10 concurrent requests, and a 27x throughput gain cutting tick latency from 30-90s to 2-4s. A SOTA engine tunes GDP and friction so population survives 2,000 ticks, settling at 96 within a 60-100 band. Every run is seeded, same hash, verified across 500-plus tests and 27 scenarios. This is not a chatbot; it is a microscope built from five theories, Adler, Freud, Marx, Maslow, Descartes, a civilization that explains itself the instant it is asked to govern itself.