play
From the Meridians Wiki · Public · Maintained · joint
Play (the room)
Episode — The immersive, Stageplay-led presentation of branch story. A catch-up Episode embodies accepted branch/canon material and introduces no new Will. A live Episode wraps a bounded Scenario, opening orders, plans, commitments, trades, or other moves through roleplay before governed resolution and staged consequence.
Episode delivery — One prepared Stageplay presented at a declared point in an ongoing episodic story. It may be prepared ahead, but must be checked against accepted branch state before release. Its ending may close an embodied scene or hand the story to Will through a bounded decision window.
Episode posture — Whether an Episode is catch-up or live. The posture changes
where control enters, not the visual medium, Stageplay language, or authority model.
Wrapped Scenario — A bounded Scenario instance presented through a live Episode. The participant roleplays a situated character and makes a critical story choice without needing to see the card-game layer; legal moves, resolution, provenance, and branch authority remain identical to direct Scenario.
Seat-shaped Episode — A rendition composed for one perspective from shared public beats and private beats reachable by that seat's role, knowledge, memory, permissions, and tools. Two seats may receive different but compatible Episodes over one branch situation. Private context remains private through Scenario resolution unless a later authorized public consequence reveals it.
Story puzzle — A consequential situation grounded in the Domain: the player must understand a rule, relationship, route, financial flow, information regime, institution, or incentive well enough to direct an order or plan under partial knowledge. It may admit several coherent moves. It is neither an object lock, an arbitrary riddle, nor a detached comprehension quiz.
Explanatory instrument — A bounded map, ledger, graph, dossier, infographic, board, portrait, or cinematic view used within a delivery to make a complex relation and its possible moves legible. It carries the current information boundary; it is not a parallel world surface.
Lobby — An optional social threshold around staged play. Humans and agents may meet, exchange partial information or resources, trade, organize, form a party, select a passage, or observe consequences. A lobby does not make multiplayer mandatory and does not own canonical World state.
Current product direction: Scenario is the shared decision substrate. It may be played directly as the fast abstract card RPG or wrapped inside a live Episode so the player encounters roleplay and consequential choices instead of explicit cards. World is later deployed from the story and assets accumulated on a sufficiently mature branch; replayability remains open. The detailed vocabulary below remains the contract for shipped Scenario unless explicitly adopted.
Fixed Episode — A catch-up audiovisual rendition created from authorised Domain state. It is fixed for an audience: playback introduces no new Will and never mutates canon.
Stageplay — The direction, production, and interaction pipeline for Episodes. It composes authorised state, reusable assets, performance, sound, strategic puzzles, explanatory views, and optional decision hand-offs. Its blocking and edit do not mutate canon, and a prepared continuation must not assume that unresolved moves have succeeded.
Authored Will / narrative Will — Shorthand for the character choices and consequences already fixed in authorised story material. Stageplay embodies that agency through performance and direction; it does not claim that playback introduced new live Will or that Will became a stored force. Live plural Will begins when several human or agent seats can make new choices through the action boundary.
Textual World prototype — An abstract, primarily text-and-graph form of World used to test continuation, per-seat context, partial information, tools, resources, cadence, and plural Will without waiting for complete audiovisual production. It is directional prototyping, not a shipped World product. Its state and action contracts may later combine with the visual identity and game-view grammar proved by Stageplay.
Experience modality — One way of acting in or reading the same Scenario or World: card table, graph mandate, governed context injection, text-led roleplay, spatial sprite presence, tactical/economic instrument, directed set-piece, or fixed Episode. A modality changes affordances and presentation; it does not create a new canonical state. The list is directional beyond the shipped Scenario forms.
Experience transition — A governed handoff between modalities that preserves Domain/branch/World identity, revision, perspective, event lineage, and control posture. A set-piece must declare its return to live presence. A transition is not itself permission to mutate the World or expose hidden information.
World — A semi-permanent, inhabitable artifact deployed from a sufficiently developed story branch. It inherits cast, places, relationships, information, resources, rules, history, open Fate, and production assets from an exact origin. It may begin as a tour and later support persistent activity or economies. Replay, reset, forking, comparison, and MMO scale remain open. See the World concept contract.
World origin — The immutable Domain, branch, revision, configuration, and asset lineage from which a World is deployed. World activity does not silently rewrite that source.
Will — The open choice a human or agent makes from the context and tools available to its seat. Will is not a stored force or personality score; the selected MCP call is its concrete expression.
The application layer. Where the substrate is used — by a writers' room, in sessions, on a regular cadence. The room is two things at once: a role-play simulator for acting a world from the inside, and a table read for deciding where the story goes. (Earlier docs called this the War Room — kept as a legacy alias for the strategic / non-fiction mode; lead with the writers'-room framing.)
The loop — Model, Capture, Scenario — Playing Meridians is one loop in three beats on a shared substrate (rule of three). Model reads the premise or material you already have and makes it playable — extraction into a typed domain. Capture (formerly Calibration) records the world's canon over time and keeps it coherent as the story grows, per-perspective by design: every seat sharpens its own Stream on its own open questions as priors accumulate. It is not belief-pooling — what is shared is the board, not the belief — which is exactly what you must not collapse when a seat plays an antagonist who should surprise you (its seat keeps deliberately hostile priors). Priors arrive two ways: live, each member tends their own seat through the full interface; dark, they drop observations into the room's encrypted Signal number and the director routes each into the seat it came from. A sparse seat just holds a wider, less-settled stance — never folded into another's. Scenario plays the chapters that haven't been written — the breadth game, the bulk of the room: only the moves the room commits to or holds at high certainty advance the story; the contested space is played forward over the Readings through board / card / information dynamics — a prototype of the story, to explore where it could go. A committed stance is a soft prior, not a constraint — contested threads are protected and play-throughs carry a divergence directive, so Scenario explores the possibility space instead of re-enacting the prior. Scenario's first shipped form is the card table. Every play-through is a branch the substrate keeps, so alternate realities can be compared across branches rather than collapsed. (An earlier framing had a third human beat, "Review" — a post-play audit plus a per-decision causal audit called "Butterfly". Both were retired as loop phases; the generated slide-deck survives as Onboarding decks.)
Room (War Room) — A perspective-based, role-played, information-asymmetric game on the substrate. The application; the table you convene to play a world live. A sufficiently primed SWF corpus deploys directly into one. Players take seats (each bound to a character or the narrator vantage), hold private logs, signal with cards, negotiate, commit to where a scene goes. Cards signal intent in public; private logs hold actual intent; the substrate keeps the bible. The board is a shorthand for the lived world — for now an information-asymmetry card game driven by how players play their variables and signal, the experience deliberately built for high feedback loops, thinking periods, and board-game play (a shared board on the console, a private hand per seat, drafting decisions). "War Room" survives as the legacy / strategic-mode alias. Still a work in progress.
Perspective — A seat. Bound to an entity (character / location / artifact) or the general narrator vantage; the world itself can take a seat (a storm is a location seat, an off-page institution a faction seat — external events are just a seat's plays). Held by one or more Members, driven by an Agent, or left empty to be AI-inferred. Each perspective accumulates Streams. Playing a scene from a perspective is the core act — inhabit a character, survey it in its own continuity, act the beat from inside it. "Seat" and "perspective" are interchangeable; seat for the play register, perspective for the data model.
Member — A human participant in the room. Exactly one holds the Director (Director) role (the single master device); the rest hold seats. Carries the routing keys that bind it across doors — a session token for live (the ngrok tunnel) and a Signal number/UUID for dark capture.
Agent — An AI scene partner: a persona (a preset personality or a custom prompt) that drives a perspective's Streams in place of a human — augmenting its suggestions, intuitions, and the continuation of its priors. Lets a thin-rostered room stay whole, puts AI seats beside humans so a scene partner always pushes back, and at the limit runs an autonomous table.
Scenario — The live practice (the fast side of the Vision thesis; formerly "Playback"): the room convened and played in real time — the table read, made playable, where a scene is acted forward against AI scene partners. The player-facing loop is Brief → Prepare → Commit → Consequence: Brief reads the scene from one seat; Prepare opens Streams and tests Priors; Commit backs a move with scarce conviction; Consequence reveals conflicts, makes the causal account inspectable, plays the Arc forward, and attributes realized Fate movement as Impact plus neutral External. The engine retains the structural phase identifiers read → write → play → resolution → graph → arc → scoring → game-theory; they are persistence and protocol terms, not competing interface language. Computer play, joined /scenario clients, Showdown, optional Graph and Decision Matrix phases, travel, chat, and the conviction economy are shipped. A table selects one authored mode discipline — Game Theory (counterplay), Writers' Room (collaborative story construction), War Room (concealed operations), Sandbox (possibility exploration), Mafia (social deduction), or Poker (signal and risk). Fine-tuning changes the mechanics without erasing that discipline: it continues to ground generated cards, agent decisions, table talk, and phase guidance. Chat remains a concurrent channel through Prepare and Commit rather than becoming a blocking phase; agents deliberate from three explicit layers — the seat's private Perspective continuity, the operating Agent's stable persona, and the public conversation — while location talk remains scoped to its current place and round. A game is about intent made visible — seats show or hide commitments to coordinate or mislead, and the engine plays them forward into a kept branch. It is the fast twin of Capture: same Threads, stances, and resolver, run in a session instead of over weeks. Scenario is a pluggable layer over the shared resolver, so other live formats can ride the same substrate later. Capitalised "Scenario" is the live practice / game; lowercase "conviction" is the resource it runs on — priced by improbability (−ln p, scale 20), it banks but decays (α = 5/6 ≈ 0.83, hoard ceiling 150). The full Scenario studio is a laptop/desktop surface; joined players also have a focused phone surface for the same live table.
External — The neutral Scenario ledger for unassigned Fate: realized movement beyond the seats’ known actions. External is neither favorable nor adverse and is not a competing player; each seat may interpret its bearing as luck, misfortune, or circumstance. It can later be resolved into other agency, conditions, events, or systems where canon supports that account. Until then, External keeps the remainder visible without inventing an attribution or treating it as evidence of a hidden force. See World Spirit for the reflective lens behind this boundary.
Access model — Console / Live (ngrok tunnel) / Dark (Signal) — How a room runs across devices. The console is the Director's computer (the master device) that holds the substrate, drives the shared screen, and is the single source of truth — no sync, no merge, no conflict. It runs an always-on Electron host: the instance stays up continuously, so members reach it any time, not only during a convened session. Non-Directors reach it two ways. Live (ngrok tunnel): the always-on host keeps an ngrok tunnel up continuously — a stable public HTTPS URL with a real cert — non-Directors open it on any phone (QR, no app install) and get the full interface (same substrate, board, live state), read-write, PIN-gated, Director elevated. Not a crippled mobile view — the real interface. An always-on tunnel is persistent public exposure (TLS terminates at ngrok's edge), so application-layer auth is the perimeter — two-stage pairing, the daemon rejects everything unauthenticated but the rate-limited pairing endpoint; the tunnel is Director-revocable, and fully-private tables run LAN-only (no tunnel). (ngrok is chosen for the stable reserved URL an always-on host needs — an ephemeral per-session quick tunnel would change address on every restart.) Dark: when the host is genuinely off, the always-open capture layer stays available for members to drop priors any time. Signal is the sovereign / team capture channel (the consumer rung uses a Telegram bot instead — reach over sovereignty; see the consumer growth engine) — end-to-end encrypted, security-first (dedicated room number via signal-cli, DMs not a group — a group would leak every seat's beliefs — quarantined one-way ingress, bare admissibility ack). No central server — the tunnel is third-party, Signal is a normal chat, the Director machine is the always-on host. The console holds the truth; live you join it over the tunnel, dark you feed it through Signal.
Sovereign master / Client — Two roles a running instance can hold, decided by origin (getSyncRole). A sovereign master is the authoritative owner of its own browser-local, encrypted substrate — every standalone instance is one: localhost, the Electron host, and a visitor who opens the canonical web app (meridians.global, a Vercel deploy) and gets onboarded into their own sovereign world in their own browser. Sovereign-by-default: a fresh browser becomes its own master and runs Director onboarding, no server account. A client is an instance that joined another host — it reached the app through a host's tunnel origin, so it renders from the master's synced state and never holds the credentials. Being a sovereign master (owning your data) is distinct from being a tunnel host (running the persistent broker + ngrok agent so clients can join): only a persistent local/Electron process can host; a serverless web visitor is sovereign but never hosts.
Director — The single person who owns the domain: the showrunner / director who hosts and curates a domain. Keeps the bible safe (the substrate is one encrypted .meridian file, backed up by the owner, optionally copied to an in-app online drive), convenes the sessions, refines the domain, and holds the canon. May be a lead inside the studio / room or a facilitator on our side. The role is the curation-and-commit layer and the gatekeeper: runs the always-on Electron host that keeps the instance and its ngrok tunnel up (shares the stable URL), and decides which inputs graduate into the canon — reviewing the Signal capture inbox when the instance comes back online, curating, and committing what earns it. Holds elevated privileges while other members have full read-write over the tunnel. Everyone else joins as a seat. Local-first by default; data responsibility is the owner's. (director is the underlying code/role term.) Federation & custody: a domain is federated — every member is a tributary feeding canon into one shared domain, and the director is the confluence holding the single canonical copy. That custody concentrates a room's collective creative work in one place, so the seat is a position of real trust, coveted and gated — coveted because it sits at the apex of the ladder and the engine trains whoever holds it (the lead sharpens fastest), gated because the same custody could be misused (the role is earned through play, never granted cold). Data sovereignty cuts both ways: it walls the room's IP off from outside extraction and binds the lead to the people who trusted them. A director delegates to admins — trusted concurrent operators who help run and refine the live world (the master local instance supports multiple concurrent admins/leads) without owning it — while the master instance and its sovereign local data stay with the lead. One source of truth, many hands, one custodian answerable for it; sovereignty does not divide.
Access hierarchy (Guest → Viewer → Contributor → Admin → Director) — The graded permission lattice within a room, by capability (distinct from the growth ladder below, which is the onboarding journey; this is what you can do at any moment). Five tiers: Guest — game access only (plays Scenario / the live room, no bible access); Viewer — bible read; Contributor — bible read + own-stream write (tends their own per-seat Priors / Stream); Admin — full bible read/write, the director's delegated concurrent operator (the master local instance supports multiple concurrent admins/leads); Director — sovereign: owns the master instance + sovereign local data, grants/revokes every tier below, holds final commit authority, answers for the whole. The gate rises tier by tier toward the seat that carries the world's trust. Everyone below Admin has no authority over the host's canon; one source of truth, many hands, one custodian.
Growth ladder (Player → Contributor → Director) — The growth model: bottom-up through the network of creators, not a top-down enterprise sales motion. Three rungs, each a deepening relationship to one domain. The player on-ramp is the card game — a creator meets Meridians by sitting down to a game of Scenario (low-friction, social, all they carry is Brief → Prepare → Commit → Consequence). Membership is the Signal chat — they join the room's back-channel and start contributing canon to the domain, becoming a Contributor. Ownership is becoming the Director — introduced to the engine, building their own domain, running their own table, recruiting the next ring. Play earns membership, membership earns ownership, owners make more players: a positive feedback loop, a network that grows itself. The director's job at every rung is to keep the room social and generative — that, not a pitch, is what moves a creator up a rung. Meridians is a networking tool for directors: running domains is how a lead builds and deepens a network — collaborative worldbuilding / role-play / simulation as a creative, social relationship-building experience — and the quality of the domains and the play is the value creation that attracts and holds the network. The first buyers are worldbuilding teams with real budgets (studios, showrunners, dev and game/animation teams); the indie creator volume tier rides the same ladder beneath them. Replaces the earlier "single-player forecasting tool" and "boutique strategy practice" self-descriptions.
Invitation — The networking funnel that grows the room from outside it. Anyone seated — member or guest, on mobile or desktop — can share a public invite link (the /join route, reachable only while the host's tunnel is live); the recipient fills a short form (email, name, mobile — no password) and the request funnels to the master's Invitations queue, where the Director admits it as a member. The link carries who shared it (?by=…) for attribution, so invitations double as the seated players' own way to grow the network for us — a person met anywhere, remote or in-room, routed to the one custodian. The invited user can later check whether they've been admitted. (Distinct from a guest pass, which is a time-limited seat-scoped token for one live /scenario seat; an invitation is a request to become a member.)
Workshop — The core growth motion and the network's entry event: a free gathering that takes a premise and makes it playable. People come to play a domain (a session of Scenario on a freshly-built domain) and, in playing, generate ideas for new and more detailed domains — so a workshop both onboards players and seeds domains. It's the social front door of the growth ladder (Player → Contributor → Director): free, social, domain-seeding, leaving a substrate and relationships behind rather than a one-off. (Distinct from the deferred studio Pilot, which a grown room can land later.)
Card — An intent signal played in the open. Either AI-dealt (from the substrate's read of available moves for the seat) or custom-authored (a bespoke beat, disclosure, or side-channel proposal). Both count as intent signals. AI-dealt cards keep the game honest to the world's possibility space; custom cards keep it honest to the player's real intuition for where the scene goes.
Private Log — An entity's hidden state, actual intent, secret information. Every seat carries one. Cards are the channel between the private log and the public game state — the player chooses to disclose, leak, or never reveal.
Public Summary — The global game-state tracker. What the table sees; the aggregate of what entities have chosen to expose. Reads at a glance via the two display surfaces: the Board (a board-game-style map with nested maps) and the Graph (the raw substrate — nodes and edges) underneath. These are the only two map types; switching between them is re-projection, not rebuilding.
Surfaces — The workspace clusters its instruments by role. Mind holds the reasoning surfaces: Complex, Theme, Graph, Readings, Threads, Phase, Decision, and Opinion. Complex is the convergent dashboard; Theme is its specialist view of evidence-backed English thematic analysis. Neither replaces the independently inspectable reasoning sources they bring together. Base holds canonical world state and topology; Content holds scene renderings; Station holds what the Program delivers; Signals holds the machine's evidence and belief flow.
Decision Matrix — The per-scene game-theoretic decomposition: a sequence of decisions (1- or 2-player), each scored with integer stake deltas, driving Nash analysis, ELO, and the player archetypes. Full vocabulary in Strategic play (game theory) below. Also called incentive structures.
Stakes — The cost layer optionally attached to plays. Off by default in the creative room — a writers' room isn't trying to beat itself; the default is perspective-divergence and branching, not "who won the round." Where switched on, three settings:
- Fictional — ELO, leaderboards, the satisfaction of a calibrated call (a fictional layer for play, opt-in).
- Reality-anchored — stakes pooled on real-world questions, with prize pools (a non-fiction / simulation-mode surface).
- Real — actual trades, hedges, or positions recorded as commitments (non-fiction mode only, legal sign-off).
When on, Stakes turn Scenario into skin-in-the-game play. These belong to the supported non-fiction / simulation modes and the opt-in fictional layer; the headline creative-room experience leaves them off.
Recall — When a situation lands, the engine asks have we been here before? — matching it against the cumulative branch history by meaning (cosine over scene summary embeddings, see Embeddings), weighted toward recent play. Grounded in recognition-primed decision science (Klein): under pressure experts recognise and retrieve a played move rather than weigh options. A situation counts as recalled only if the match clears a similarity threshold and the matched branch was room-played, fresh, and carries a committed response — cosine alone is geometric approximation, not recognition. High recall infers preparedness: the room has met this and knows the move. (Also the IR sense — of the relevant cases, how many we retrieved.)
Prior Knowledge & Foresight — The two readiness metrics (demoted from headline North Stars to the supported non-fiction / simulation mode under the story-prototyping repositioning — still true, no longer the lead). They measure how much played material a branch carries, scene by scene. Both read the one cosine matrix over scene summary embeddings pooled across all branches: split by play order, earlier scenes feed Prior Knowledge (backward — have we been here before?), later scenes feed Foresight (forward — do we already foresee what's coming?). Each similarity is passed through a steep logistic centred on a match boundary (s₀ = 0.80, κ = 22) and the strength is the mean of the top-K (K = 5) — so a 0.90 match counts ~9× a 0.70, five strong matches sway far more than one, and mediocrity collapses to zero. A backward match on another branch — a future the room actually played — is weighted up (×1.4). Advisory and continuous, computed from the simulation; their reality-anchored proof is Recall Share (below). The geometry advises; reality scores. (Retired from the shipped app — the Experience scoring surfaces and the scene-summary similarity matrix behind them were removed; kept here as vision vocabulary only.)
Recall Share — The reality-anchored metric of the supported non-fiction / simulation mode (demoted from a headline metric under the story-prototyping repositioning): the external proof of Prior Knowledge & Foresight, anchored in reality rather than the simulation. Of the consequential events that actually landed this period, the share the room had already played (recalled per Recall, above) — "eleven hit the desk, nine were recalled." Because the events are real and checkable, an outsider can verify it; the practice never grades itself. Reported per branch and per arc, raw and recency-weighted, 0–100. Wider play and regular capture are the INPUT (the way training is the input to match performance), never the score itself. Recognition is auditable now; accuracy (calibration on confirmed thread resolutions, Brier) is the slower second layer — see Believability.
Believability — Weighting an idea by the track record of the seat behind it, not its seniority — an idea meritocracy you can audit (Dalio's Principles is the cultural antecedent — the track record it asserted but never had a number for). Two claims, kept distinct so neither overreaches: recognition is auditable now — a seat's committed stance is frozen before the outcome, its played branches stay on file with reasoning attached, and Recall shows when the room had met a situation before; accuracy is the slower layer — recognition proves a seat played, not that its judgement was good, so as threads resolve observably their confirmed outcomes (walled in software from the believed merge) are scored by a strictly proper rule (Brier). Believability earned by calibration is the destination; auditable recognition is what ships first. Full calibration infrastructure is deliberately deferred — the hardest part to operationalise honestly — but the believed/confirmed wall and confirmed resolutions on fast-resolving threads are cheap and real, so the practice consults the record, never grading itself.
Session — One meeting of the room. Typically one to two hours for the weekly cadence (an episode's beats, a sequence, an act), two to four hours for the monthly cadence (a season arc, a world's mythology, the shape of the whole story). A session is the unit of practice.
Cadence — The declared rhythm under which a game accepts consequential play and advances. Scenario may convene as a live session. Every World is turn-based and may declare active hours, off-hours, weekends, and other breaks so invested humans and idle agents can share a humane clock. The exact phases and mapping between wall time, turns, and fictional time belong to the Domain or preset.
Tick / turn allocation — A seat's declared opportunity to decide or act inside a World cadence. Different entities may receive different tick eligibility, action budgets, cooldowns, queue priority, concurrency, or tool latency according to System rules. Rate can therefore become a capability, cost, or economic resource, but it must remain distinct from accidental compute and scheduler delay. The record keeps allocation, use, waiting, expiration, and acceleration legible because iteration speed can compound into power over the world.
Practice — The institution of regular sessions. The compounding loop. The fiftieth weekly session is qualitatively different from the tenth because the domain underneath has absorbed fifty cycles of play, fifty curated canon drops, fifty branches explored. A domain is a habit before it's a product. Practice is what makes the domain deepen into something no tool can ship cold.
Onboarding decks — Generated slide decks over a slice of the substrate — a room's history and the key decisions that shaped it — that update a newcomer's priors when they join a team, move between teams, or are introduced to a new room. The lightweight onboarding surface (no session to convene), and it regenerates as the substrate moves so it stays current. Pushed to members' phones, it doubles as the knowledge-transfer surface; resolved threads and pivotal decisions make natural deck beats. (Survives from the retired Review phase, where it was the deck-delivery mechanism.)