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.

The purpose of the model is to explore the influence of actor behaviour, combined with environment and business model design, on the survival rates of Industrial Symbiosis Networks (ISN), and the cash flows of the agents. We define an ISN to be robust, when it is able to run for 10 years, without falling apart due to leaving agents.

The model simulates the implementation of local waste exchange collaborations for compost production, through the ISN implementation stages of awareness, planning, negotiation, implementation, and evaluation.

One central firm plays the role of waste processor in a local composting initiative. This firm negotiates with other firms to become a supplier of their organic residual streams. The waste suppliers in the model can decide to join the initiative, or to have the waste brought to the external waste incinerator. The focal point of the model are the company-level interactions during the implementation or ending of synergies.

…

This is the model for a paper that is based on a simulation model, programmed in Netlogo, that demonstrates changes in market structure that occur as marginal costs, demand, and barriers to entry change. Students predict and observe market structure changes in terms of number of firms, market concentration, market price and quantity, and average marginal costs, profits, and markups across the market as firms innovate. By adjusting the demand growth and barriers to entry, students can […]

This model studies the effect of the agents’ adaptive expectation on cooperation frequency in the prisoner’s dilemma game in complex networks from an agent based approach. The model is implemented in Repast simphony 1.2.

The HUMan Impact on LANDscapes (HUMLAND) 2.0.0 is an enhanced version of HUMLAND 1.0.0, developed to track and quantify the intensity of various impacts on landscapes at a continental scale. The model is designed to identify the most influential factors in the transformation of interglacial vegetation, with a particular focus on the burning practices of hunter-gatherers. HUMLAND 2.0.0 incorporates a wide range of spatial datasets as both inputs and targets (expected modelling results) for simulations across Last Interglacial (~130,000–116,000 BP) and Early Holocene (~11,700–8,000 BP).

Model explains both the final state and the dynamics of the development process of the wine sector in the Małopolska region in Poland. Model admits heterogeneous agents (regular farms,large and small vineyards).

Reducing packaging waste is a critical challenge that requires organizations to collaborate within circular ecosystems, considering social, economic, and technical variables like decision-making behavior, material prices, and available technologies. Agent-Based Modeling (ABM) offers a valuable methodology for understanding these complex dynamics. In our research, we have developed an ABM to explore circular ecosystems’ potential in reducing packaging waste, using a case study of the Dutch food packaging ecosystem. The model incorporates three types of agents—beverage producers, packaging producers, and waste treaters—who can form closed-loop recycling systems.

Beverage Producer Agents: These agents represent the beverage company divided into five types based on packaging formats: cans, PET bottles, glass bottles, cartons, and bag-in-boxes. Each producer has specific packaging demands based on product volume, type, weight, and reuse potential. They select packaging suppliers annually, guided by deterministic decision styles: bargaining (seeking the lowest price) or problem-solving (prioritizing high recycled content).

Packaging Producer Agents: These agents are responsible for creating packaging using either recycled or virgin materials. The model assumes a mix of monopolistic and competitive market situations, with agents calculating annual material needs. Decision styles influence their choices: bargaining agents compare recycled and virgin material costs, while problem-solving agents prioritize maximum recycled content. They calculate recycled content in packaging and set prices accordingly, ensuring all produced packaging is sold within or outside the model.

…

This multi-model (i.e. a model composed of interacting submodels) is a multi-level representation of a collective motion phenomenon. It was designed to study the impact of the mutual influences between individuals and groups in collective motion.

The model explores how two types of information - social (in the form of pheromone trails) and private (in the form of route memories) affect ant colony level foraging in a variable enviroment.

The first simple movement models used unbiased and uncorrelated random walks (RW). In such models of movement, the direction of the movement is totally independent of the previous movement direction. In other words, at each time step the direction, in which an individual is moving is completely random. This process is referred to as a Brownian motion.
On the other hand, in correlated random walks (CRW) the choice of the movement directions depends on the direction of the previous movement. At each time step, the movement direction has a tendency to point in the same direction as the previous one. This movement model fits well observational movement data for many animal species.

The presented agent based model simulated the movement of the agents as a correlated random walk (CRW). The turning angle at each time step follows the Von Mises distribution with a ϰ of 10. The closer ϰ gets to zero, the closer the Von Mises distribution becomes uniform. The larger ϰ gets, the more the Von Mises distribution approaches a normal distribution concentrated around the mean (0°).
In this script the turning angles (following the Von Mises distribution) are generated based on the the instructions from N. I. Fisher 2011.
This model is implemented in Javascript and can be used as a building block for more complex agent based models that would rely on describing the movement of individuals with CRW.