Prorok world model is a definition of a domain. It's a formal description of what can exist in a simulation, how it behaves and how state evolves over time.

The purpose of a world model is to serve as ground truth for higher level reasoning, be it by a human or an AI agent.

A world model brings together:

• Ontology: entities and components, generally and the types of state a simulation can hold, and
• Logic: behaviors, services, and the rules that drive transition from one state to the next

Together, these define a reusable blueprint. One world model can power many worlds, meaning concrete simulation instances with different starting conditions, scenarios or random seeds.

Executable

Crucially, Prorok worlds are executable, meaning we perform explicit computation to transition to next states based on the available logic.

This gives us strong auditability. Every run can be logged and replayed, so we can track results all the way back to individual behaviors and interactions.

It also means we're focusing on laying out the correct assumptions about the simpler building blocks up-front, meaning we don't require large amounts of data for training, at least for most use-cases.

When we create a simulation, we start from a world model, instantiating a world instance to explore and track outcomes.

Dynamic

One important aspect of Prorok world models is that they are dynamic, meaning there's capability to mutate the model at runtime.

As each world instance is explicitly driven by the underlying model with both ontological and behavioral schema, the ability to mutate the model at will opens up the door to supporting both intricately self-modifying systems and nested simulation.

This is especially useful for models tracking complex intelligent agentic behaviors, with agents being able to perform sub-simulations of slices of the larger world to drive their own planning and decisions.