World — the Domain as a place you can wander

# World — the Domain as a place you can wander

Status: **direction; working concept contract, not an implementation contract.**

This specification owns what is currently stable about World and what remains open. Current runtime truth
remains in the code and generated MCP manual. The [human experience
contract](../../knowledge/product/positioning/human-practice.md) owns how Scenario and Episode develop the
stories from which Worlds may later be deployed.

Founder roadmap: [World, Factory, and latent-character roadmap](../../sources/founder/2026-09-02-2026-09-02-world-factory-roadmap-founder-note.md).

## The working identity

**A World is the third medium: a Domain's state rendered as a place, held at a chosen moment, and explored
freeform.** Where a Scenario is decided and an Experience is read, a World is *wandered* — the reader moves
a cursor over the map rather than advancing through an authored order, enters the locations the story
built, and speaks with the characters who were there.

It is a **static artifact of world state with a live interaction layer**, not a clock that advances merely
because someone entered it. The world is pinned to a coordinate on a branch, and the tip of that branch is
only its most recent coordinate, not its only one. Traversal and conversation may generate while a visitor
wanders, but they do not change the selected moment or source branch. That branch supplies the ontology,
history, characters, places, artifacts, institutions, rules, relationships, information, resources, open
Fate, and reusable audiovisual assets already established through Scenario and Episode play. Bounded
generation may fill gaps, but it must extend that inherited ontology rather than invent an unrelated
setting.

A World may initially be toured, observed, or revisited. A richer World may later support situated human
and agent activity, financial and informational economies, social organization, and MMO-like presence.
Replay, reset, comparison, and continuous simulation remain possible configurations, not the settled
centre of the concept. World is deliberately subordinate to the useful Domain and the Scenario/Episode
play loop: Meridians does not need a persistent World or MMO economy to deliver value.

> **Domain defines. Branch preserves. Scenario resolves. Episode embodies. World inherits and lives.**
>
> **You decide in a Scenario, read an Experience, and wander a World.**

### The cost shape is the opposite of an Experience

An Experience is capital: generated once during production, then served, so cost per reader is hosting.
A World inverts that — the map and its assets are prepared, but **conversation and traversal generate at
the moment of interaction**, so cost scales with how long and how widely a visitor wanders. The
"a prepared world does not generate while you read it" property in
[Positioning](../../knowledge/product/positioning/promise.md) belongs to Experience and must never be
restated for World. World spend is an ordinary metered lane and is bounded by the same instance ceilings
as every other paid call.

## The moment cursor

A World's defining control is its clock. Entering a World selects a **moment** — an `asOf` coordinate on a
branch — and the world is composed *at* that moment: who was where, what had already happened, which
artifacts existed and who held them, which threads were still open, and what each character could know.
Moving the cursor is time travel through the same place, not a different World.

This is not a new primitive. It is the same as-of vantage a Reading already uses to cut a fan of futures
without hindsight (`Reading.contextScope: "asof"` with `asOfEntryId`). One clock serves both: the Reading
uses it so the future is fair, the World uses it so the past is inhabitable.

## Presence is derived, not stored

Where a character was at a moment is **already recoverable from canon** and should not become a second
stored register. Canonical `Scene` records carry `locationId` and `participantIds`, and scenes are ordered,
so folding scenes up to the cursor yields each character's last known place, each location's cast, and the
traffic between them. The spatial signal derivation in `src/lib/core/graph/spatial-coherence.ts` already
counts presence at every containing level from exactly these fields.

The honest limits of that derivation, which the interface must respect rather than paper over:

- It yields **last known position**, not a trajectory. Between two scenes a character's whereabouts is an
  inference, and the World should present it as such rather than animating a walk that canon never recorded.
- Travel is implicit. A character appearing in a new location is a fact; the route they took is not.
- Off-screen characters have no position at all until the story places them. Absence of record is not
  presence at the last place seen, and the interface should be able to say "not recorded here".

If a stored movement register is ever added, it must arrive as a typed scene delta on the canonical write
path — never as state the World writes for itself.

## The map is already an artifact

The explorable surface has a shipped substrate. A `Board` is a Replicate-rendered image of one location
cluster plus the snapshot of the cluster it was generated from, with `MapLabel` entries placing each child
location at normalized `[0..1]` coordinates over that image — resolution-independent hit targets for a
cursor. `LocationTopology` supplies containment and portals, so which places adjoin which is a graph rather
than an art decision, and cluster drift is already detectable through the board's `signature`.

What is missing is the traversal layer over it: selecting a location, descending into a cluster, populating
a place with the cast present at the cursor's moment, and entering conversation from there.

### Factory supplies the visible world

The visual map is not a second asset system. World draws from the same production library as Experience:
approved location plates and variants, character sprites and appearance sets, artifacts, maps, effects,
sound, motion, and interface material. The directional **Factory** workspace in
[visual-novel production](../visual-novel/production.md#12-factory--the-production-control-surface) owns the
coverage, lineage, approval, and repair of that library. The existing Media Drive remains the file and
lightweight generation/gallery seam; Factory is the production control surface that says which entity or
moment an asset belongs to, what reference and style produced it, where it is valid, and what is still
missing.

World consumes an exact asset package at deployment. A later asset regeneration does not silently recast
an already deployed World: it creates a new asset revision that a new World origin or explicit update may
select.

## Situated roleplay

Conversation inside a World is Domain chat with two boundaries added: **where** and **when**. A character
spoken to at a moment answers from what that character knew then — their location, their memory, their
information position, the threads open to them — and not from canon at the branch tip. The information
boundary is the same authorization contract that Scenario seats already enforce; the World applies it to a
place and a time rather than to a seat at a table.

Situated roleplay consumes Acting context. When an entity has an Entity Interpretation Graph, it resolves
the branch-and-moment State and Dynamics bindings active at that coordinate. It composes those provisional
models with the character's available knowledge,
relationships, present state, constraints, and tools to produce a Behaviour Reading. The reading supplies
competing tendencies, expressions, and micro-movements; it does not reveal a true interior or choose an
action. See [Acting and Entity Interpretation](../latent-model/README.md#9-activation-into-behaviour).

This makes character change legible without making it magical or fixed. New accepted history may become
evidence, but a new interpretation does not require evidence divergence: several supported graphs may read
the same record differently. An explicit refresh creates a new graph snapshot; a later branch coordinate
may bind it. The visible change can be gradual or a step change, including a new expression or
appearance coverage request to Factory. Earlier moments retain the model and assets they used. Different
branches may therefore carry different evidence-grounded interpretations of the same continuing entity
without either interpretation becoming canonical psychology.

A visitor may also take a character's part and move as them. Two invariants hold regardless:

- **Roleplay never writes canon.** A conversation inside a World is a projection, exactly as a playthrough
  of an Experience is. If a session produces something worth keeping, it is promoted through Scenario or a
  Stream on the ordinary governed path — never merged because it was said.
- **The moment does not advance because someone spoke.** A World visit leaves the branch where it found it.

## Story cultivation before deployment

Scenario and Episode develop the main story or several divergent stories at active branch heads:

```text
direct Scenario play → governed branch move → optional Episode rendition

live Episode framing → wrapped Scenario decision/resolution → branch move → staged continuation
```

Direct Scenario exposes the abstract card and rapid-simulation grammar. A live Episode wraps a bounded
Scenario instance in Stageplay, roleplay, situated information, and consequential choices, so a player can
participate without knowingly entering a card game. A catch-up Episode embodies already accepted history.

These paths are interoperable, not rival writers. Both accumulate the story facts and production material
that make a later World coherent.

## Origin, branch, and canon

A deployed World pins an exact origin:

```text
Domain id + source branch id + preserved revision + World configuration + asset lineage
```

The origin is immutable even if activity within the World later changes. A World may originate from an
official canon branch or a preserved alternate timeline. Several Worlds may therefore embody different
branches of the same fictional Domain, and many Worlds may share one origin with different configurations.

Creating or inhabiting a World never silently advances or rewrites its source branch. Any World event
proposed back into story canon requires an explicit governed action and preserved provenance. Whether a
Domain designates exactly one official canon branch, and how that designation can change, remain open
policy—not an inference from branch order or popularity. Founding a new Domain from a scene of this one
(a **seed**) is that governed action made concrete, and consequence then flows down lineage only
([world-tree](../world-tree/README.md#4-seeding--how-one-narrative-begets-another)).

## Inherited and completed space

A World should inherit what the story has genuinely earned:

- named characters, Units, locations, artifacts, institutions, and their identities;
- System rules, permissions, clocks, economies, and tool affordances;
- relationships, ownership, obligations, Frames, Control, and resource positions;
- information distribution, memory, belief, secrecy, and discoverability;
- history, unresolved Fate, tensions, opportunities, and consequences;
- reusable actor, location, prop, effect, sound, motion, map, and interface assets;
- branch-and-moment Acting bindings used for behaviour and expression, with the evidence and
  interpretation lineage kept distinct from canonical entity facts;
- exact source, branch, revision, configuration, and adaptation provenance.

Episodes will develop some places deeply while leaving connective or ordinary areas unstaged. Generated
completion may fill those gaps only through an explicit World-building operation constrained by inherited
ontology, style, geography, history, and rights. Generated completion is attributable derived material,
not retroactive evidence that the source story established it.

Asset reuse across branches must preserve semantic identity and revision lineage. Shared assets may descend
from a common ancestor; branch-specific changes must not silently overwrite one another. The precise asset
alignment and merge model remains open.

## Presence, seats, and activity

A playable seat may bind to a character, location, artifact, institution, or another Domain-authorized
entity. Each seat receives context bounded by its graph position, memory, information, resources,
permissions, and tools. Canonical truth is not common knowledge.

A World may support, in increasing order of commitment:

- a tour or guided traversal of story-developed places;
- perspective-shaped observation of preserved history;
- bounded roleplay, quests, work, trade, or social activity;
- persistent human and agent seats;
- financial or informational economies and organizing systems;
- lobby-based or MMO-scale participation.

No item in that sequence is mandatory, and it is not yet a promotion ladder. A tourable World may be a
complete and valuable form.

## The five forces, interpretation, and Will

A World inherits the [pentad](../pentad/README.md) of its origin Domain — five peer forces, three stored
and two derived:

- **World** embodies — people, places, artifacts, resources, relationships, information, and state. *Stored.*
- **System** constrains — rules, institutions, clocks, economies, permissions, and tools. *Stored.*
- **Fate** remains open — unresolved possibility, pressure, and bearing. *Stored.*
- **Theme** recurs — meaning read across the record; never a writer. *Derived.*
- **Causality** connects — what can lead to what, traced as **Flags**: turning points that signal Causality
  into action. A **Thread closure** is a Flag that also settles a Fate thread; it is a derived projection.
  *Derived.*
- **Will** makes the move — a human or agent chooses through the canonical action boundary. Not a force.

**Acting is a first-class entity-scale interpretive capability** and a post-hoc interpretation of System;
an Entity Interpretation Graph is its selectable State or Dynamics context artifact. First-class does not
make Acting a force or canonical interior truth. Will is never stored as a force. It is attributable authority
exercised when a human or agent creates, revises, or selects an interpretation and when it makes a move.
Acting reads may recur as the World changes, but generation and binding remain manually initiated; a World
does not reinterpret its actors on its own.
Presentation, including Stageplay, never resolves an action merely by depicting it.

## Information, tools, and economies

Information propagation is foundational: reach, delay, fidelity, credibility, secrecy, distortion, and
decay may all matter. A World's tool ecology declares which capabilities exist, who may use them, what they
cost, and how they compose.

Information, artifacts, access, time, money, credit, reputation, ownership, labour, and tool capability may
support trade, finance, coalition, institution, and organizing mechanics where the Domain has earned them.
These systems are a plausible route from a toured story artifact to a livable MMO-like society. They are
not promised for every World, and fictional economies do not become real financial claims.

## Projection and Stageplay

World state and history are canonical; cameras, Stories, dashboards, scores, and Stageplays are
projections. A World may be experienced through a side-scrolling stage, map, ledger, graph, dossier,
spectator account, tour, or another Domain-specific instrument.

The World may reuse production assets accumulated through Episodes. It is not itself a long Episode, and
Stageplay does not become its state owner. Conversely, activity inside a World may later be rendered as an
Episode if authority and provenance permit.

## Roadmap — scout, make, read, wander, learn

The roadmap is one production loop over one Domain, not separate VN and World pipelines. It is grounded in
the current [VN experiment audit](../../sources/research/2026-09-02-2026-09-02-vn-experiment-audit.md):
the lab already has a deterministic catch-up reader, several VN-first fixed-video proofs, wrapped Scenario
routes, and candidate CG sequences. It is still `rehearsal-ready`, with no held-out trials and substantial
approval, cast-continuity, acting, audio, and audience-review gaps. Those experiments are evidence for what
to integrate next; they are not shipped product state.

```text
Fate Exploration scouts branch terrain (MWS grows the tree between readings)
→ promoted candidates declare production coverage
→ Factory builds and approves reusable entity/place/moment assets
→ Experience freezes route + latent-model + asset bindings for reading
→ World freezes moment + latent-model + asset bindings for wandering
→ governed consequence creates new evidence
→ explicit latent reinterpretation and Factory repair prepare the next coordinate
```

Three workstreams can advance without pretending they are independent products:

- **production spine:** Factory → prepared Experience → static World tour → situated interaction;
- **discovery spine:** Fate Exploration → promoted route → Factory coverage demand;
- **character spine:** evidence profile → Entity Interpretation Graph → Behaviour Reading → expression demand.

The gates below determine integration order. Research may run ahead of a gate, but a later surface may not
claim the capability until its dependency passes.

### What the gates are for — three forms, one substrate

The roadmap is not an ontology programme. Its purpose is to make **worlds worth experiencing, being inside,
and growing from** — the same kept world entered three ways. Each gate below exists because it moves one
of the three forms toward its evidence; a gate that does not is research, not roadmap. The first released
unit is the [training world](../training-worlds/README.md): an hour-long original Experience with a
canonical good route, divert and rejoin, and a reading given back afterwards. The forms, their evidence, and where they sit in the strategy:

| Form | What it must become | Gates that build it | Evidence it produces | Owning leaf |
|---|---|---|---|---|
| **Experience (VN)** | a prepared world read at book-like cost, whose second seat is a second book and whose record is the reader's | 0 · 1 · 2 · 3 · 4 · 5 | one Experience played end to end; completion and second-seat rate | [04](../../knowledge/product/strategy/04-the-market.md#what-is-sold-each-with-one-gate), [12](../../knowledge/product/strategy/12-the-practice-monopoly.md) |
| **Training worlds for agents** | a pinned world × task × seat × interval a policy can be run through, with the log kept as the result | 1 · 5 · 7 · 8 | someone runs their own agent through a pinned world and changes what they let it do; one measure repeatable, discriminative, useful | [13](../../knowledge/product/strategy/13-consequential-test-worlds.md) |
| **Pocket dimension (populated World, VNMMO)** | a small kept world a few people enter as its characters and step back into, its history held between visits; solitary practice made social, bounded before persistent ([01](../../knowledge/product/strategy/01-the-north-star.md#pocket-dimensions-with-care)); its first form is not concurrent play but a family of timeline-dated VNs released into a shared multiverse ([world-tree](../world-tree/README.md)) | 6 · 7 · 8, then the open list below | a bounded multi-seat loop makes the debrief richer, not merely the world busier | [04](../../knowledge/product/strategy/04-the-market.md#what-is-sold-each-with-one-gate) |

Read across the rows, the same gates appear under more than one form: that is the point. Factory lineage
(1) and latent interpretation (5) are what make a world *consistent enough to be re-entered*, which is at
once what a returning reader, a policy under test, and a second player all require. Situated interaction
(7) is a static tour with a seat that can commit — the first configuration in which a human wanders, an
agent is tested, and a second Will arrives, all through the same governed hand-off. Held-out production
(8) is the first honest proof for all three that the world holds up under circumstances it was not
prepared for.

Two loops run beside the gates and are owned elsewhere. **Meridians World Search (MWS)**
([branch refinement](../branch-refinement/README.md)) grows the branch tree every form is cut from; under
the pocket dimension it becomes a shared tree with a value per seat, pricing joint futures and choosing
for none. **Latent traits** ([spec](../latent-traits/README.md)) keeps each person's record across every
world they play and steers the next hour toward them; inside a populated World that record stays the
person's own — other seats see acts and consequences, never a reading.

The order is deliberate: **Experience first**, because it is the cheapest form to prove and the one whose
record starts soonest; **training worlds for agents beside it**, because a run is a Scenario with the log
kept; **the pocket dimension last**, because it is the most expensive and least evidenced form, and because it
is populated by people who practised in training worlds first.

Read as priorities ([specs map](../README.md#priorities-2026-09-21)): the **causal simulator** — a kept
world whose pathways MWS has grown and priced, entered through the VN and, as direction, built from
a person's own situation by interview — is the value and what Meridians monopolises; gates 0–5 build it.
**Micro sonic analysis** ([sonic](../sonic/README.md)) is the value-add on its felt grain. The **VNMMO** is
the stretch goal on a separate roadmap: a persistent universe that relies on the same simulator and runs it more automated —
MWS as a shared tree, worlds constructed and kept without a hand on each fork — and it is built
beside the simulator, never instead of it. Gates 6–8 and the open list are its road; nothing in them is
allowed to slow gates 0–5.

### Gate 0 — consolidate the existing VN proof

Freeze Episode 006's twelve-moment reader passage and the strongest VN-first cuts from Episodes 008, 009,
011, and 013. Review them at phone scale with sound on and off. Add the exact editorial clock, restrained
audio, and only the two motivated transitions already named by dailies; do not create more CGs first. The
gate passes when the reader contract and editorial grammar survive human review without weakening their
presentation-only authority or digest-pinned lineage.

### Gate 1 — Factory as production memory and asset lineage

Establish the reusable supply before either new medium depends on it: protected identity references,
versioned sprite and appearance sets, expressions, plates and variants, style/model/reference stamps,
branch-and-time applicability, coverage gaps, estimates, and explicit approval. Fate Exploration may create
draft thumbnails and a coverage manifest; it may not autonomously promote expensive assets.

The first increment is a rebuildable semantic projection over the lab's existing catalogs, asset usage,
dependency closure, receipts, dailies, approval, cost, demand, and readiness records. It is not a new byte
store or a gallery-first tab. The gate passes when one character across two branches reports reusable,
missing, incompatible, stale, and orphaned coverage and freezes an exact approved package without silently
changing any dependency. Serialize the shared paid-generation budget writer before Factory schedules
concurrent acquisition.

### Gate 2 — prepared Experience MVP

Produce one small branching visual novel from a promoted fan. Every passage pins its Script versions,
seat projection, active latent-model snapshots, and exact asset revisions. Branch-specific behaviour and
expression may differ where evidence or an explicit evidence-grounded interpretation differs, while the
reader remains a projection and writes nothing to canon.

Promote the proven reader controls and VN grammar through the permanent Stageplay and export seams rather
than copying the lab renderer. Start catch-up only; the composed decision remains inert until a governed
live Scenario hand-off exists. The gate passes when a prepared route can be rebuilt from its package,
read outside the authoring surface, and audited back to source, consequence, character interpretation, and
asset receipts.

### Gate 3 — CG lane

Treat Episode 006's three CGs as candidate sequence evidence, not an approved family. Give CGs a
moment-scoped lineage, reference roles, approval, cost, and orphan report; spend them at irreversible deltas
or Director-approved bottlenecks. Prove contextual identity continuity, two motivated transitions, sound
on/off comprehension, and gallery unlocking before widening the provider or image count. A beautiful
isolated frame does not pass this gate.

### Gate 4 — Fate Exploration to Factory hand-off

Run the scout structurally and cheaply over a fixed budget. Promote only delta-gated, sufficiently
orthogonal futures; then emit entities, places, moments, Behaviour Reading needs, reusable coverage, gaps,
and estimated production cost to Factory. The gate passes when one promoted fan reaches the prepared
Experience without the scout generating or approving protected production assets.

### Gate 5 — latent character interpretation

Begin offline and in parallel with Gates 1–4: build evidence profiles at an as-of coordinate and generate
Entity Interpretation Graph candidates. Prepared Experience and World work pin the selected graph and
binding per branch and moment directly; Behaviour Reading activation and held-out comparison are an
unscheduled extension, not a prerequisite. If activation is ever built, it may request semantic expression
coverage from Factory but never choose an action or rewrite evidence. The extension earns runtime authority
only when one entity expresses coherently across two divergent branches and a later, independent evidence
cluster earns an explicit new snapshot rather than mutating history.

### Gate 6 — static World tour

Open one preserved branch coordinate through the existing Board and topology: move the cursor, descend into
locations, inspect the cast and artifacts present at the selected moment, and reuse the same Factory assets.
No live character generation is required to prove the spatial memory palace.

The gate passes when the tour reconstructs presence from canonical scenes, preserves the exact moment,
latent-model, and asset bindings, and introduces no stored position register or second world authority.

### Gate 7 — situated interaction

Add bounded conversation and observation at the selected location and moment. Context is seat-, place-,
time-, and latent-model-shaped; generation is metered; the session neither advances the moment nor writes
canon. A worthwhile outcome can be proposed back through Scenario or a Stream on the governed path.

### Gate 8 — held-out production and learning

After accepted consequences create sufficiently independent evidence, refresh rather than mutate the
entity's latent model. Compare the old and new Behaviour Readings, preserve contradictions, select the new
snapshot from an explicit effective coordinate, and let Factory report any newly required expression,
appearance, voice, or motion coverage. This is the first honest proof of an emergent character: continuity
is preserved while the proxy used to simulate expression becomes more adequate.

In parallel, freeze unseen Experience briefs before asset resolution and measure plan, technical,
semantic, human, audience, repair, time, and cost gates. Iteratively repaired Episodes remain valuable hill
climbs but do not count as on-demand proof. Expand from a small calibration set toward the Stage Kit's ten
held-out-trial hypothesis only after the end-to-end seam is stable.

Fate Exploration is the scout across this sequence, not its producer. It maps causal terrain, identifies
which entities, places, moments, and behavioural states a promoted route would require, and sends that
demand to Factory. Factory makes coverage; Experience and World consume it.

## Deliberately open

- whether touring, inhabitation, social activity, or economy is the first lead use;
- whether World activity is ephemeral, persistent, forkable, resettable, replayable, or continuous;
- what “mature enough” means for a branch and who authorizes deployment;
- how one official canon branch is designated among alternate timelines;
- asset inheritance, divergence, and alignment across branches;
- the exact Factory collection schema and whether its first implementation is desktop-only;
- the boundary between inherited world material and generated gap completion;
- cadence, action resolution, delegation, lifecycle, and persistence;
- how World history relates to its source branch and whether it may propose canon;
- scoring, discovery, migration, upgrades, hosting, sharding, and MMO scale;
- how a repeatable training loop joins Program orientation, Episode immersion, Scenario commitment,
  consequence, debrief, and player progression without reducing ability to generic XP;
- real-value economies, licensing, moderation, and integrity design.

The first World proof should not assume replay or character autonomy is the point. It should take one
story-developed branch coordinate, inherit a small location and cast with exact model and asset lineage,
fill one clearly labelled gap, and prove a static tour before adding bounded interaction without changing
the source branch.

Raw Markdown source