The world is not the frame.

A world larger than its current scene

A visual game engine does not confuse the existence of a world with the need to render every part of it at once. The player may inhabit a space much larger than the screen, while the engine constructs the view needed from the current position.

This distinction is one of the foundations of scale. The world can preserve more terrain, objects, processes, and history than the current frame displays. Rendering is the act of selecting and reconstructing the correct local experience from that larger whole.

Myriuna begins from the same broad systems problem in narrative space. A persistent world may contain locations, actors, relationships, obligations, institutions, private knowledge, active processes, and accumulated history. A participant in one scene does not need all of it in active context. They need the part that is true, relevant, available from their perspective, and useful for the act currently taking place.

A persistent world and its current narrative rendering are not the same object. The world may be large. The frame must be specific.

What a narrative physics engine means

Traditional engines provide an engine-level commitment to physical causality. They preserve positions, movement, collision, terrain, materials, forces, and the consequences of bodies acting upon one another.

Myriuna is being built to provide an equivalent commitment to another layer of a world: memory, knowledge, trust, obligation, loyalty, reputation, status, law, rumour, authority, and personal or institutional history.

The phrase narrative physics does not imply that people are mechanically predictable. Actors may misunderstand, conceal motives, contradict themselves, revise their beliefs, or surprise the player. The engine is intended to preserve the conditions and consequences within which that ambiguity has meaning.

A broken promise may alter trust. Changed trust may alter what someone reveals, offers, risks, or refuses. Those changes may affect a household, profession, friendship, institution, village, or political conflict. The social action becomes a durable cause rather than a line of dialogue that disappears when the scene ends.

Context needs a camera

Much discussion of AI context begins with capacity: how much information can a model hold at once? Myriuna starts from a different question:

Can the system reconstruct the right information, at the right resolution, from the right perspective, under the right authority, for the act currently taking place?

A larger context window gives a model more material to inspect. It does not independently determine what matters now, which record is current, what a particular actor may know, how much detail is useful, or what the model is allowed to change. Those remain architectural responsibilities.

Persistent world or map

Authoritative world state, history, rules, actors, relationships, and active processes

Camera and field of view

The situated participant and the act currently taking place

Visibility and occlusion

Perspective, knowledge, relevance, causality, and authority boundaries

Asset streaming

Context retrieval and progressive narrative revelation

Level of detail

Semantic resolution appropriate to distance, role, risk, and consequence

Rendered frame

A compiled, scene-ready local reality

State update

A validated consequence returned to persistent world state

Culling is not forgetting

A visual engine avoids processing everything that cannot affect the current frame. A narrative engine requires an equivalent discipline.

Information may be excluded because it is spatially distant, outside the participant’s perspective, unknown to the actor, temporally obsolete, causally disconnected from the current situation, outside the model’s authority, or simply irrelevant to the present act.

That exclusion does not erase the information. It determines what becomes present now.

Absent from the frame is not absent from the world.

Narrative streaming and semantic resolution

A persistent event does not need to appear at full detail the moment it begins. A distant war may first become visible through higher prices, then travellers’ reports, displaced families, changed patrols, shortages, and eventually direct involvement.

The underlying process can persist while its narrative representation becomes more detailed as it approaches the participant’s life. The world does not begin changing because the player looked at it. The player gains increasing access to a world that was already changing.

The same object can also be represented at several deliberate levels of narrative detail:

1. World-scale invariant: what remains broadly true.
2. Regional summary: the institutions, conflicts, routes, and pressures that matter at regional scale.
3. Local operating state: how those pressures currently affect a particular place.
4. Actor-specific knowledge: what one participant knows, believes, misunderstands, or suspects.
5. Scene-ready detail: the people, obligations, risks, and immediate conditions needed for the present act.
6. Evidentiary lineage: the deeper records required when the system must audit, explain, or update the object.

The objective is not minimum context and it is not maximum context. It is sufficient, status-aware, perspective-correct detail for the present act, with clear routes to retrieve deeper history when the situation genuinely requires it.

The engine owns reality. AI renders it.

The simulation and its expression must remain connected but distinct. Persistent world state determines what exists and what has happened. Narrative rendering determines how that reality becomes locally experienceable.

How the studio intends to use these practices

Myriuna is not being approached as one large storyteller carrying an entire universe inside a prompt. The studio is separating the work into concrete system responsibilities that can be implemented, tested, and inspected.

• Authoritative state outside the model: persistent facts, histories, actors, rules, and relationships remain owned by the engine.
• Viewpoint-aware context compilation: the system reconstructs a bounded local reality for a particular participant and act.
• Progressive retrieval: deeper evidence and higher-resolution state are loaded when the situation requires them.
• Expressive generation: AI performs dialogue, description, interpretation, atmosphere, and proposed action inside the compiled frame.
• Validation before reintegration: proposed consequences are checked against authority, rules, current state, and permitted change.
• Recoverable lineage: accepted changes return to persistent state with enough context to support future scenes, review, and correction.

These responsibilities are at different stages of development. Public claims should therefore remain tied to visible evidence: what has been implemented, what is currently being tested, and what remains planned.

The claim, and its limits

Myriuna is not claiming that narrative state behaves identically to pixels, textures, geometry, or a visual renderer. The claim is that both belong to the same broad class of systems problem:

How does a large persistent reality produce a coherent local experience for a limited participant under finite processing constraints?

Visual engines learned that scale does not come from presenting everything simultaneously. Scale comes from preserving more than is currently shown and reconstructing the correct experience on demand.

Myriuna is being built to test that principle in narrative space. The aim is not to load an entire universe into the thinker. It is to build the universe as persistent state, render the part currently being lived, and let validated consequences return to the whole.