A context layer is a structured specification that defines a world: its rules, its entities, its constraints. Load one into a language model and the model becomes a simulation of that world. Every input updates state. Every response is the next frame.
The layer is swappable. Write a different specification, get a different simulation. The engine underneath is the same. One architecture runs strategy games, kingdom management, crisis training, market modeling, and courtroom simulations. The context layer is the only thing that changes.
A context layer specifies everything the simulation needs: what can be measured, what forces act on outcomes, how volatile the world is, what stages a turn moves through, who the autonomous actors are, what discrete conditions exist, and how the output is presented.
Starfall Legion defines five ships, their captains, weapons systems, crew counts, morale tracking, combat phases, and a display grid. Dominion defines a kingdom, its treasury, military, populace, seasons, and event systems. Same engine. Different layers. Different worlds.
Starfall Legion. Five ships, 4,074 crew. Procedural encounters. Boarding parties, railgun volleys, fighter wings. A mirror fleet built to be unbeatable. The question is how long you last.
Dominion. Medieval kingdom management. Treasury, people, military, reputation. Every choice trades them against each other.
Pleb Life. Ancient Rome from the bottom. Survival when you own nothing.
King’s Dilemma. Rule a kingdom where every option has a cost. There are no right answers.
Four games. Four context layers. One engine.
The engine is domain-agnostic. Any setting where decisions have consequences and state evolves over time is a candidate for simulation.
Crisis response. A hospital overflows, supply chains break, decisions cascade. Train responders in a simulation that tracks every variable and produces a full state history for debrief.
Courtroom simulation. Depositions, cross-examination, objections. Witness agents with consistent testimony and motivation across the proceeding.
Market modeling. Trading agents that react and adapt. Inject shocks. Observe how cascades propagate through the system.
Negotiation training. Counterparty agents with hidden interests and constraints. Run the same deal a hundred ways. See where each path leads.
Multi-agent advisory. CFO, COO, Risk Officer. Each sees a tailored projection of the same world. Each proposes. The engine reconciles into a single outcome.
Organizational dynamics. Restructuring, policy changes, leadership transitions. Model how decisions propagate through a system of agents with their own goals.
Prairie Engine is a simulation engine for language models. No training, no fine-tuning, no custom models. One text file turns any frontier model into a persistent, stateful simulation.
Write a specification. Load it. Run a world.