Five models, shared experimental methods.

Explore how simple interactions produce complex behavior. Each world runs in your browser, with a sandbox, challenges, and a hypothesis lab for comparing controlled interventions.

Artificial life and evolution

origin

What changes when survival depends on both behavior and the environment?

Organisms with evolving neural controllers compete for food and reproduce in a shared environment.

Open world

Collective behavior

swarm

When do local interactions produce coordinated motion?

Agents respond to nearby neighbors through alignment, cohesion, and separation rules.

Open world

Pattern formation

morph

How do reaction and diffusion rates change a pattern?

A Gray-Scott reaction-diffusion system forms spots, stripes, and other spatial patterns.

Open world

Agent-based markets

market

How do agent strategies affect prices and volatility?

Trading agents with different strategies interact through an order book in a simplified market.

Open world

Opinion dynamics

social

Which interactions lead to consensus or polarization?

Opinions evolve as agents interact across a network with adjustable topology and influence.

Open world

Run an experiment

The hypothesis lab uses the same procedure across all five worlds.

  1. Define the comparison

    State a hypothesis and choose a control and treatment. Change a parameter and select the outcomes to measure.

  2. Repeat with seeded runs

    Run multiple replicates so the comparison accounts for variation between runs. Reusing a seed reproduces the same simulation.

  3. Inspect the result

    Compare the measured outcomes, effect sizes, and corrected significance tests. The result describes the model under those conditions.

Scientific reasoning benchmark

Petri-bench gives model agents hidden interventions to investigate through a limited number of experiments. It scores the answer and whether the conclusion is supported by an isolating experiment.

Programmatic access

The MCP server exposes the same simulation engines to agents for deterministic, headless experiments.

npx -y petri-labs-mcp
Package and instructions