model-7

mymodel-5-1

mymodel-5

mymodel-4

asdf

mymodel 2
mymodel 2

MASTOC-LLM extends the classic Multi-Agent System Tragedy of the Commons (MASTOC) model by replacing hard-coded behavioral rules with autonomous decision-making powered by large language models (LLMs). Three heterogeneous agents manage herds of cows on a shared grassland commons. Each tick, an agent receives a structured prompt describing current resource levels, its own herd size, peer behavior, and — optionally — a rolling memory of recent rounds and messages from neighboring agents. The LLM returns a stocking decision (add, remove, or hold cows) together with a natural-language rationale and, when communication is enabled, a short message to broadcast to peers.

The model is designed to test whether LLM agents spontaneously develop Ostrom-style common-pool resource governance (mutual monitoring, graduated sanctions, graduated rule revision) or instead fall into identifiable failure modes. Preliminary experiments with Claude Haiku 4.5, GPT-5.4-mini, and DeepSeek R1:32b have revealed four recurring collapse patterns — Cooperative Paralysis, Defection Cascade, Overshoot-Panic, and Hybrid Architecture Failure — whose onset timing is sensitive to memory length, inter-agent communication, and the post-training alignment approach of the underlying model.

MASTOC-LLM is intended as a laboratory for generative agent-based modelling (GABM) methodology: it provides a clean, well-understood commons baseline against which LLM behavioral hypotheses can be systematically tested and compared across models, parameter sweeps, and alignment regimes.

This repository contains the Python implementation of an agent-based model investigating how localized boundary-crossing dynamics generate large-scale connectivity in structured multi-attractor landscapes.

Agents evolve in a continuous two-dimensional environment composed of attractor basins. A fraction of agents exhibits exploratory higher-mobility dynamics, while the remaining agents remain locally constrained. The model analyzes how localized configurational transitions accumulate into transition networks that progressively integrate the explored state space.

The repository includes:

…

Exploring learned cooperation, coevolution and free-riding. Learning is achieved through Multi-Agent Deep Reinforcement Learning (MADRL) in an ecological environment. The environment emits no other than sparse reproduction rewards. No reward shaping, no explicit cooperation signal.

Interest-based compound economies generate monotonically increasing wealth inequality through multiplicative accumulation dynamics, yet the conditions under which gift-based reciprocal exchange outperforms such systems in collective well-being remain unquantified. We present Zensei Wago (全生和合), a seven-layer agent-based model comparing a Gift Resource Circulation (GRC) economy with a Compound Interest Circulation (CIC) economy under identical initial conditions. Across N = 5000 Monte Carlo replications (T = 700 ticks, N = 100 agents), GRC produced significantly higher collective resonance than CIC (p < 0.001, Cohen’s d = +0.171), above a critical prosocial threshold pm ≈ 0.698. Cohen’s d grows monotonically with duration — d = +1.943 at T = 1500 and d = +4.126 at T = 3000 — driven primarily by structural collapse of CIC resonance as inequality exceeds a critical Gini threshold (G > 0.333), while GRC resonance remains stable. The gift mechanism further decouples collective well-being from distributional outcomes, generating resonance through relational quality rather than material redistribution. Network topology analysis across seven configurations — combining a Watts-Strogatz rewiring sweep and a T = 1500 longitudinal replication — reveals that ring topology maximises GRC advantage (d = +1.17), that most topology-dependent reversals are transient (sparse and small-world both transition to significantly positive by T = 1500), and that a critical rewiring threshold of p ≈ 0.10–0.20 separates GRC-advantaged from GRC-disadvantaged network configurations. Scale-free networks remain persistently adverse (d = -7.24*), requiring structural redesign for gift-economy viability.