DOCS / CORE CONCEPTS Docs overview

Core concepts

The Counterworld mental model.

A Counterworld world is a running software system whose state changes through agent actions, simulated users and time.

One world, four connected layers

The layers are separate enough to configure independently, but they operate on one shared state. An order created by a simulated buyer is the same order the agent later reads through the API.

01SystemsEntities, relations and valid mutations
02UsersActors grounded in world state
03HistoryPrior records, events and commitments
04ClockWhen scheduled work and reactions occur
Shared-state rule: systems, users and time never maintain parallel versions of reality. Every change lands in the same world state.

Definition and instance are different

A definition describes what the world can contain and how it behaves. An instance is one populated, running realization of that definition.

World definitionCommerce systems, rules, actor types
World instanceMaple & Moss, 42,184 orders, Day 731
CheckpointExact state at Day 731, 09:00
Definition
Versioned systems, schemas, mutations, actor types and evaluators.
Instance
Concrete records, identities, event queue, history and current clock.
Checkpoint
An immutable reference to an instance at an exact moment.
Fork
A new running instance that begins from a checkpoint.

State records what is true. Events record what happened.

State answers questions such as “is this invoice paid?” The event history explains how it became paid, who changed it and which side effects followed.

  • State: current entities, relationships, permissions and pending schedules.
  • Events: ordered mutations with actor, time, inputs and effects.
  • History: coherent prior state and events generated before the agent starts.

Actors and time create work independently of the agent

Simulated buyers, customers, store admins, sales reps, support agents, employees, approvers and operators observe the current world, make decisions and act through the same valid mutations. The clock determines when they react.

ObserveRecords, prior events and current conditions
DecideGoals, constraints and behavior parameters
ActCreate orders, replies, returns or escalations

Evaluation compares resulting worlds

The primary output is not a response string. It is the world state produced by the agent after other actors and time have responded.

Evaluators can check immediate invariants, delayed consequences and aggregate outcomes such as retention, backlog, margin or policy violations.