Date Compiled: 2026-04-28
Type: concept — factory-methodology
Related Questions: shared-cognitive-architecture, multi-agent-memory, agent-communication-protocol, shared-state
Related Articles: superada-multi-agent-architecture, superada-enterprise-crew, him-model, Kelly router, multi-agent-pipeline, memory-system
Source Attribution: raw/superada/epic-chat-world-model.md · compiled/sources/superada-multi-agent-architecture.md

World Model — Shared Cognitive Architecture for Multi-Agent Systems¶

What Is the World Model?¶

The World Model is the Enterprise Crew's shared cognitive substrate: a single world.json file that serves as the authoritative source of truth for agent state, context, and coordination. It is the answer to the question: "How does a signal from Henry reach the right agents at the right time?"

Before the World Model, each agent reasoned from isolated files. There was no shared reality — each agent maintained its own picture of the world, and coordination required the orchestrator (Ada) to manually translate and relay context between agents.

The World Model changed this by establishing one canonical state file that all agents can read and write to.

The Problem It Solved¶

Before the World Model, Henry and Ada ran autonomy architecture simulations across 6 candidate designs, scoring each on:
- Context sharing
- Failure resilience
- Cascade risk

The conclusion: the problem was not orchestration patterns. The problem was that agents had no shared reality. Each agent reasoned from its own isolated files. There was no single source of truth.

The result was a system that looked like multi-agent but behaved like a collection of isolated services that happened to share a task queue.

The Architecture¶

The World Model has three components:

1. world.json — The Shared State File¶

A structured JSON file that every agent in the crew can read and write to. It contains:
- Agent state and health
- Active task ownership
- Context propagation signals
- Priority and escalation state

2. Propagation Model — Signal Routing¶

How a signal from Henry reaches the right agents:

Henry (intent) → Ada (orchestrator) → world.json (state) → relevant agents (execution)

The propagation model defines rules for how information written to world.json flows to the agents that need it, based on role, current load, and availability.

3. pi-research Tracks — Stress Testing¶

Three parallel research tracks that stress-test the design by running the architecture through adversarial scenarios:
- Scale testing (up to 50 simulated agents)
- Failure mode testing (what happens when an agent goes offline mid-task)
- Cascade risk testing (does a failure in one agent propagate to others)

Performance Metrics¶

The World Model was validated against the following metrics:

The composite score of 8.85 reflects both effectiveness (E) and dependability (D) across the six candidate architectures tested.

Why It Matters¶

Before the World Model, agent-to-agent context transfer required Ada to manually relay information — she had to know what each agent knew and what each agent needed. This made Ada a bottleneck.

After the World Model, agents read their context from a shared file. Ada no longer needs to be the middleman for every piece of context — she writes to world.json once, and agents read what they need.

This decouples the orchestrator from the cognitive load of maintaining each agent's context.

Connection to Kelly Router¶

The Kelly router in dark-factory-kb uses a different approach to context: sub-agent sessions are spawned with explicit context injection, and the router uses multi-agent-pipeline patterns to route work. The Enterprise Crew's World Model is a peer concept — both solve the "how do agents share state" problem — but the SuperAda approach uses a shared file (world.json) while the Kelly router uses session inheritance and spawn-time context injection.

Both are valid; they reflect different tradeoffs between consistency (shared file) and isolation (session inheritance).

Connection to Memory System¶

The memory-system concept in dark-factory-kb describes a layered memory architecture (session → project → global). The World Model is conceptually adjacent — it serves as the "global" layer for the Enterprise Crew, the shared file that persists across all agent sessions.

Sources: epic-chat-world-model.md (raw/superada/), superada-multi-agent-architecture.md (compiled/sources/)

Related Articles¶

Kelly router, him-model, memory-system, multi-agent-pipeline, superada-enterprise-crew, superada-multi-agent-architecture