How we work together

阅读中文版

Meet-U's current onboarding path is a collaboration, not a public upload box. The IP owner supplies or approves the human creative foundation. The team and its tools help make that intent explicit, testable, and reviewable without quietly taking over canon or character agency.

The assisted pilot workflow

1. Confirm fit, scope, and a safe channel

We begin with a non-confidential summary and one focused pilot question. Before protected source material moves, the participants, intended use, access, retention expectations, and rights contact need an appropriate private process. Public issues and public chats are not intake channels for confidential or licensed IP.

Use Prepare a pilot to organize the conversation without exposing source material prematurely.

2. Shape a human-authored dramatic brief

Together, we narrow the work to one character, one remote-confidant relationship, and one bounded dilemma. The brief identifies character values, contradiction, voice, knowledge, boundaries, faithful choices, world outcomes, secrets, return moments, and visual intent.

The point is not to generate more plot. It is to preserve the choices that make this character and story specifically yours.

3. Turn approved meaning into a structured draft

An implementation collaborator may normalize the brief into the current Creator Source Package, connect its references, and surface gaps that the compiler can check. This technical representation separates character choice, world fact, audience influence, disclosure, dialogue obligation, and visual evidence.

The structured draft is still a draft. A plausible completion from an AI tool is not canon unless the Creator reviews and adopts it.

4. Review and correct with the Creator

The Creator reviews the complete meaning, not merely a polished sample chat. Corrections should cover:

• identity, voice, values, contradiction, goals, and prohibited portrayals;
• which audience influence is plausible and which character choices remain autonomous;
• objective world outcomes and the causal separation between audience advice, character decision, and result;
• knowledge, secrecy, disclosure, exact lines, and facts that must stay out of text or images;
• visual identity, source provenance, outcome composition, and prohibited depictions; and
• behavioral examples that distinguish faithful variation from a character or authority violation.

Requested corrections should be reflected in a new exact source revision so review does not float separately from the material it approved.

5. Compile and evaluate a local candidate

The current compiler validates the structured package and deterministically produces local candidate artifacts. The team can check their reproducibility and review their structured meaning against the Creator's intent. Compilation does not execute acceptance cases or prove that every authored story mechanism is available in the current preview.

The latest local owner preview can exercise ordinary conversation, a minimal turn trace, and two manual checkpoints. The first commits a bounded character decision, deterministic outcome, and wait state. Current application code uses the second to publish a validated proactive return, one package-approved static outcome visual, a structured Journal, and a public-safe causal timeline. This fixed slice does not provide ambient scheduling, external notification, live image generation, or automatic activation for a newly compiled package.

A successful compile is an engineering result. It does not prove rights, creative approval, asset approval, behavior quality, publication, or activation. Those claims must never be inferred from a green build.

6. Make any release decision separately

A future tester or public release needs separate evidence and explicit authority bound to the exact content, assets, runtime configuration, audience, and intended use. Corrections, expiry, revocation, and rollback must remain possible without rewriting history.

The current public documentation does not offer that release workflow. The present assisted pilot stops at local, non-activatable candidate work unless a separate authorized process establishes a broader scope.

What AI assistance may do

Within the agreed scope, AI-assisted tools may help:

• organize source notes into candidate canon, character, relationship, arc, disclosure, dialogue, visual, and evaluation sections;
• identify contradictions, missing references, ambiguous authority, or a choice that would weaken character agency;
• propose alternate wording or structured drafts for human review;
• check a package against typed constraints and known portrayal boundaries; and
• compare candidate behavior with Creator-authored evaluation intent.

Generated suggestions remain proposals. If the Creator deliberately adopts a suggestion, it should become reviewed source with clear provenance rather than remain hidden model output.

What AI assistance must not decide

AI assistance does not have authority to:

• declare ownership, licensing, consent, approval, or permitted use;
• replace the human-authored dramatic foundation with bulk generated plot;
• change canon, character identity, moral boundaries, or disclosure policy without Creator review;
• force the character to obey audience instructions or treat a message as a world event;
• invent objective outcomes or reveal private facts outside the approved possibility space;
• approve an asset, imitate a voice or likeness, choose training use, or publish a release; or
• pool Creator IP or private audience-character history for general training without explicit, appropriately scoped consent.

Control points for the IP owner

Creator control should remain visible at every meaningful boundary:

Control point	What the Creator reviews or decides
Intake scope	Which material may be used, by whom, for what purpose, and for how long
Canon and portrayal	Identity, voice, values, relationships, knowledge, boundaries, and prohibited portrayals
Dramatic space	The dilemma, faithful decisions, objective outcomes, secrets, and causal structure
Audience relationship	What the audience may influence and what they can never control or do inside the world
Visual direction	Source authority, identity anchors, permitted transformation, scene intent, and forbidden depictions
Evaluation	Which examples demonstrate fidelity, unacceptable behavior, or unsafe disclosure
Exact revision	The precise source and assets covered by a correction or approval
Release scope	Audience, locale, runtime, intended use, expiry, revocation, and rollback authority

The current tools cover only part of this lifecycle. Missing product UI does not make a control optional; it means the pilot must stop before that claim can honestly be made.

Current status and boundaries

Today:

• onboarding is private and assisted; there is no public IP upload or self-service Creator Portal, and no approved private-intake contact has been published;
• the supported runtime scope is local owner review with synthetic input, not tester, consumer, public, or production delivery;
• current application code implements two manual checkpoints through a validated proactive return, approved static outcome visual, structured Journal, and public-safe timeline; it does not provide ambient scheduling, external notification, or live image generation;
• the compiler produces deterministic artifacts marked non-activatable;
• existing images can be bound to a pilot, but live image generation and automatic asset approval are not part of the authoring compiler;
• one package version selects en-US or zh-CN as its default locale; the compiler does not translate the full package; and
• current fact guards support bounded withholding, but not a durable multi-turn trust/probe ladder for secret disclosure.

See current capabilities and limitations for the precise implementation boundary.

Handoff to the technical path

IP owners can remain in the assisted creative review path. A technical collaborator implementing an agreed pilot should continue with:

1. the semantic guide to preserve Creator, character, audience, and world authority;
2. the quickstart to construct and compile the current English package;
3. the field reference for exact schema syntax; and
4. compilation and lifecycle before describing any candidate as reviewed, approved, or releasable.

YAML is the current implementation team's interface. It is not the creative qualification for becoming a Meet-U IP partner.