CORE PILLARS

Foundations

CPS is currently a bounded prototype built around local-first interaction, structured memory, traceable propagation, durable persistence, and CPS Studio observability. The system is designed to separate interpretation from state. Language models may help interpret conversations and generate responses, but CPS preserves the underlying structure: source, ownership, confidence, privacy, lineage, memory records, and history. This makes the project more than a conversation interface. It is an experimental continuity system for artificial worlds, built to make social memory visible, inspectable, and eventually reusable.

CPS is not presented here as a finished commercial platform, a production-scale society simulator, or a validated scientific instrument. It is an active prototype and creative technology investigation. The immediate priority is a small, stable, inspectable demonstration before broader scaling.

THREE OPERATIONAL PILLARS

How the system holds together

The public surface is simple: capture what matters, let it move, and preserve the fiction of a world inhabited by people instead of chat agents.

01

Durable Memory Records

Conversations are distilled into durable narrative records so the system can retain facts, identity cues, and preferences across sessions.

FCT Facts World knowledge, reports, and specific claims.
IDT Identities Who someone is, what they want, and how they are known.
PRF Preferences Affinity, aversion, and emotional triggers that shape later recall.
  • Extracts structured memory from dialogue and reflection.
  • Stores world knowledge, personal detail, and preference signals.
  • Gives later interactions a persistent factual backbone.

If you tell a character something shareable today, another resident may later encounter its consequences through a traceable social record.

02

Controlled Propagation

Information does not stop with the original exchange. It travels through the simulation through recipient-owned processing and social handoff.

  • NPCs continue to process and share information beyond the player's presence.
  • Propagation follows social links instead of global broadcast logic.
  • Recipient-owned recall ensures that characters learn from each other as well as from the player.
  • Traceable lineage reveals how a memory mutates as it moves.

The focus is on relationship-shaped spread that can be inspected for lineage and mutation tracking.

03

Authenticity Guardrails

NPCs are constrained to behave like residents of the simulation. The system protects tone, persona continuity, and conversational plausibility.

Identity profiles, state of mind, and expertise steer what a character notices, what they ignore, and how they phrase what they know.

  • Responses stay in-character rather than slipping into assistant behavior.
  • Identity and state guide what a character notices and how they speak.
  • Continuity reduces repetitive greetings and out-of-scene narration.
PRINCIPLE_01

AI is the interpreter

EXTRACTION / PROPAGATION / RECALL

Language models parse and reformulate narrative information.
The system owns the state.

PRINCIPLE_02

AI is not the source of truth

CANONICAL DATABASE / TIMELINE CONSISTENCY

Every memory is anchored to structured persistence so continuity survives beyond a single exchange.

TECHNICAL ASSURANCE_01

Canonical persistence

The system writes durable records into a shared canonical database so continuity does not depend on a model session remaining alive.

TECHNICAL ASSURANCE_02

Structured synthesis

Extraction is organized as structured synthesis rather than vague summarization, which gives later recall and propagation a cleaner substrate.