V6 Diagram Atlas — Egbe Companions

This atlas provides the cross-cutting visual models for Egbe's persistent agentic companions. It covers cognition, player communication, durable biography, life-cycle state, runtime composition, subsystem ownership, stewardship, and release evidence. Focused pages and current runtime/evaluation evidence remain authoritative.

Agent life and cognition landscape#

flowchart TB World[Orun authoritative world] --> Perception[Perception and current context] Biography[(Ori identity biography memory and consent)] --> Recall[Governed recall] Perception --> Moirai[Moirai deterministic life tick] Recall --> Cognition[Cognition and grounded planning] Moirai --> Cognition Guidance[Player stewardship through VAC] --> Cognition Cognition --> Policy[Autonomy welfare safety and capability policy] Policy --> Action[Committed world or social action] Policy --> Refusal[Refusal delay or alternative] Action --> World Action --> Biography World --> Clio[Clio narrative derivation] Biography --> Clio Biography --> Foundry[Foundry endings and legacy] Biography --> Aye[Aye cross-game incarnation] Action --> Evidence[(Replay provenance eval and audit evidence)]

Probabilistic cognition is enclosed by deterministic state, durable identity, and governed effects. Narrative is derived from life events; it is not the canonical life ledger.

Steward communication sequence#

sequenceDiagram actor Player participant VAC as Voice and text boundary participant Ori as Ori session participant Memory as Biography and memory participant Grounding as Context and grounding participant Policy as Autonomy and welfare policy participant World as Authoritative world Player->>VAC: Advice question or request VAC->>Ori: Typed intent modality and disclosure Ori->>Memory: Recall consented relevant continuity Ori->>Grounding: Resolve current world and source context Ori->>Policy: Propose response or action alt refuses or redirects Policy-->>VAC: Reason boundary and safe alternative else conversational response Policy-->>Ori: Scoped response permission Ori-->>VAC: Grounded response with uncertainty else governed action Policy->>World: Authorized versioned intent World-->>Ori: Durable outcome or failure receipt Ori->>Memory: Record consented event and provenance Ori-->>VAC: Honest outcome and recovery end VAC-->>Player: Accessible text voice and embodiment

Understanding a request does not obligate compliance. Recall, grounding, autonomy policy, world authority, durable outcome, and visible response remain separate review points.

Durable Ori identity and life records#

erDiagram ORI ||--o{ BIOGRAPHY_REVISION : has ORI ||--o{ MEMORY_RECORD : remembers ORI ||--o{ RELATIONSHIP : participates ORI ||--o{ LIFE_EVENT : experiences ORI ||--o{ GOAL : pursues WORLD ||--o{ WORLD_REVISION : versions WORLD_REVISION ||--o{ LIFE_EVENT : grounds MEMORY_RECORD }o--o{ LIFE_EVENT : references PLAYER ||--o{ STEWARDSHIP_LINK : holds ORI ||--o{ STEWARDSHIP_LINK : permits STEWARDSHIP_LINK ||--o{ CONSENT_RECORD : governs ORI ||--o{ INCARNATION : embodies REALM ||--o{ INCARNATION : hosts LIFE_EVENT ||--o{ PROVENANCE : records LIFE_EVENT ||--o{ EVAL_EVIDENCE : evaluates

Ori identity is not a mutable profile blob. Biography revisions, episodic memory, relationships, events, goals, stewardship/consent, and incarnations carry separate lifecycles and deletion/export semantics.

Ori life-cycle state#

stateDiagram-v2 [*] --> Forming Forming --> Active: identity biography and world placement valid Active --> OfflineProgressing: no observing client OfflineProgressing --> Active: bounded deterministic catch-up Active --> Incarnating: consented cross-realm transition Incarnating --> Active: target embodiment and continuity verified Incarnating --> Recovering: target failure or version conflict Recovering --> Active: prior incarnation restored Active --> Paused: welfare privacy or operator hold Paused --> Active: governed resume Active --> Ending: terminal life event accepted Ending --> Legacy: Foundry seals biography and inheritance Legacy --> Archived: retention and access policy Forming --> Withheld: required consent or safe configuration absent Withheld --> [*] Archived --> [*]

Pause, ending, legacy, archive, and deletion policy must not be conflated. Cross-game incarnation adds an embodiment; it does not fork the identity without an explicit conflict or merge model.

Runtime contract composition#

classDiagram class OriIdentity { oriId biographyVersion consentVersion } class MoiraiTick { worldRevision seed elapsedWindow } class CognitionEnvelope { perceptionRefs memoryRefs modelPolicy } class GovernedIntent { goalRef capability expectedVersion } class LifeEventReceipt { eventId worldRevision provenanceRef } OriIdentity --> CognitionEnvelope : scopes MoiraiTick --> CognitionEnvelope : triggers CognitionEnvelope --> GovernedIntent : proposes GovernedIntent --> LifeEventReceipt : may commit LifeEventReceipt --> OriIdentity : advances biography

The concrete schemas contain more detail, but these roles preserve replay: a tick and identity version produce a cognition envelope; policy admits a bounded intent; the world returns a durable event receipt.

Subsystem and ownership estate#

mindmap root((Egbe ownership)) Identity and continuity Ori biography and memory Consent stewardship and relationships Aye incarnations Foundry endings and legacy Life and cognition Moirai simulation Perception grounding and planning Agent behavior and action Clio emergent story World and embodiment Orun and shard continuum UE client and agent bodies Gateway pixel streaming and fallback VAC voice text and turn-taking Governance and operations Welfare safety and provenance Evals drift and cost tiers Accessibility localization and commerce Performance security release and recovery

The estate is a reading aid. Each branch has a focused feature and architecture home, plus evidence documents for schema evolution, recovery, threat, welfare, cost, evaluation, and release.

Stewardship journey#

journey title A healthy V6 stewardship loop section Encounter Observe Ori world state and disclosed uncertainty: 5: Player Communicate through accessible voice or text: 5: Player, VAC section Deliberate Recall consented context and grounded world facts: 5: Ori Weigh goals relationships autonomy and welfare: 5: Ori, Policy section Act Respond refuse or commit a bounded world intent: 5: Ori Record outcome provenance and continuity: 5: World, Biography section Maintain Let player inspect correct pause export or forget: 5: Player Evaluate drift welfare cost and recovery: 5: Operators

Stewardship is successful when the player can influence, understand, and recover without erasing the Ori's governed autonomy or fabricating continuity.

Evidence order#

timeline title V6 release evidence Contract : Ori identity consent biography and world-state invariants : Model tool capability and embodiment boundaries Runtime : Deterministic ticks governed cognition and durable actions : VAC world client fallback story legacy and incarnation Verification : Continuity autonomy welfare safety and replay evals : Accessibility cost scale security and recovery tests Candidate : Migration compaction restore rollout and rollback proof : Current model provider and operations evidence Supported : Named tiers clients regions capabilities and life-cycle paths : Ongoing drift incident and welfare monitoring

The evidence order is not a calendar. Every supported claim carries a current model/client tier and the life-cycle states it actually verifies.

How to use this atlas#

Use the landscape and sequence to trace cognition/action, the ER/state diagrams for identity and lifecycle reviews, the class/estate models for ownership, and the journey/timeline for welfare and release. Continue to the V6 guide, feature index, and architecture index.