MONITOR Documentation Map

Welcome to the MONITOR Agent Documentation. This structure is designed to be highly modular and dense to maximize context efficiency.

1. Product (What are we building?)

Vision & Modes: The core pitch, modes of operation, and non-goals.
Epics: The 12 core development epics and their coverage.
Ideal State & Use Cases: Examples of how the three modes should ideally operate.

Project Status: Canonical what-is-done / what-is-not, verified against code (2026-07-23).
Gap Remediation Plan: Plan addressing every verified open gap (G-1 to G-9).
GM Assistant Mode Plan: Full implementation plan for the Co-Pilot mode (supersedes G-5).
Forge Mode Plan: Full implementation plan for the World Design / authoring mode.
GM as Authority: How narration works now — the GMAgent → Narrator → Resolver pipeline, the gm_tools registry, and GMVerdict.
Retrieval Service: The single owner of embeddings + Qdrant (pinned model, model/dim guard, HyDE + rerank, nearest).
De-heuristic Principle: Why classification is semantic/LLM, never keyword tables — and what that removed.
Play & Forge Direction: Product direction for mode-first play and the authoring/playing split.
Character Templates & GM Conditioning Plan: Extends Play & Forge §5/§6 with live multi-world test evidence; adds the (previously undesigned) story-premise conditioning gap.
Ingestion Pipeline Audit: Findings + remediation plan for game-system ingestion and character-creation parsing (silent degenerate extraction, missing provenance, step_type mislabeling).
Multi-Tenancy Plan: Phase 1 of multi-user support — per-account private campaigns, public template/character-card gallery, and the auth/ownership model needed to get there; preserves the seam for a later multiplayer phase.

Scene Loop: Turn-by-turn resolution.
Story Loop: Campaign progression.
Conversation Loop: NPC dialogue logic.
World Building Loop: Collaborative creation.
Image Generation: Canon-anchored image generation — asset lifecycle, provider capabilities, scope budgets, moderation, and loop suggestions.
Ecosystem Interop: SillyTavern/RisuAI character card import/export, lorebook runtime semantics, card macros, and lorebook directives.

Fact Canon Levels: How truth is managed.
Entity Types: Archetypes vs Instances.
Graph Relationships: How Neo4j nodes connect.
Visual Identity: Structured, canon-aware appearance description — anchor shapes, precedence, incarnation isolation, and the CanonKeeper boundary.

Gameplay Examples & GM Craft: Narrative examples and design principles.
Use Case Catalog: The full list of target workflows and specifications.
E2E Testing (LLM-vs-LLM + full-stack): How to run the autonomous-GM and replay suites.