Ethical Attractors: Universal Patterns of Cooperation and Fractal Emergence in Complex Adaptive Systems

We present a theory of Ethical Attractors—detectable, stable basins of cooperation in multi-agent systems governed by local interactions. Our contribution is threefold: We formalise attractor existence in potential games observed through noisy sensors. We prove that common adaptive rules, including soft-majority imitation and Q-learning, ascend the potential and converge to the attractor with high probability. We validate the theory through large-scale simulations in an open-source codebase spanning agent-based lattice models and reinforcement learning in iterated prisoner’s dilemma environments. The observer-network framing transforms ethics from a normative aspiration into a practical control signal: by publishing coarse cooperation metrics such as mean cooperation (μ) and payoff entropy (H), institutions can locate and enlarge Ethical Attractors, reducing governance cost Key Features Exact potential-game formulation with welfare–diversity composite Ψ = Φ + κH. Observer-network framework that detects attractor entry using only noisy local signals. Asynchronous convergence bound derived from Freidlin–Wentzell metastability. Reproducible simulations: 64×64 lattices, Watts–Strogatz, Barabási–Albert, and tabular Q-learning (5 000 episodes, ε-decay). Open-source pipeline (reproduce.sh) regenerates all figures, tables, and JSON logs in a single command.