Architecture

MCOP Framework Architecture

Back to M-COP WIKI | See also: Installation-and-Quickstart | API-Reference

---

Overview

The MCOP (Meta-Cognitive Optimization Protocol) Framework implements collective intelligence through stigmergic coordination — a mechanism inspired by ant colonies where agents coordinate through environmental traces rather than direct communication.

Core Insight: Just as ant colonies coordinate via pheromone trails, AI agents in MCOP coordinate through persistent "cognitive traces" in a shared memory substrate.

The canonical expansion (since 2026-04-26) is Meta-Cognitive Optimization Protocol. Earlier documents may reference "Multi-Cognitive Optimization Protocol" or "Meta-Cognitive Operating Protocol" — these are historical variants of the same project.

---

The Core Triad

All processing in MCOP flows through three kernel components, always in sequence:

Input
  │
  ▼
[ NOVA-NEO Encoder ]  →  context tensor + entropy score
  │
  ▼
[ Stigmergy v5 ]       →  resonance score + Merkle-proof hash
  │
  ▼
[ Holographic Etch ]   →  rank-1 micro-etch + confidence delta
  │
  ▼
ProvenanceMetadata bundle (hashes, scores, lineage)

---

System Components

1. NOVA-NEO Encoder (NovaNeoEncoder)

Also known as ContextTensorEncoder.

Purpose: Converts any input (text, context string) into a deterministic, fixed-dimension numerical tensor.

Key properties:

• Deterministic — the same input always produces the same output vector
• Entropy-normalized — each encoding includes an explicit entropy/uncertainty score
• Configurable dimensionality — default 64 dimensions; adjustable via dimensions config
• Optional normalization — controlled by normalize: true/false

What it solves: Provides a reproducible, comparable numerical representation of cognitive context, enabling similarity matching downstream.

---

2. Stigmergy v5 (StigmergyV5)

Also known as SharedTraceMemoryV5.

Purpose: A shared vector-pheromone memory store that records past (input, output) pairs and retrieves the most similar past case using cosine similarity.

Key properties:

• Pheromone traces — each stored record (PheromoneTrace / MemoryTraceRecord) links an input vector to a synthesis vector plus metadata
• Cosine similarity recall — scale-invariant, O(d) similarity search over the stored trace library
• Configurable resonance threshold — resonanceThreshold: 0.4 (default); only matches above this score are returned
• Merkle-chained — every trace carries a Merkle-proof hash for tamper-evidence and distributed verification

Biological analogy: The Stigmergy component works exactly like pheromone trails in an ant colony — traces decay-free and accumulate to form the collective memory of the system.

---

3. Holographic Etch (HolographicEtch)

Also known as ChangeAuditLogger.

Purpose: An append-only change ledger that records every small confidence update (rank-1 micro-etch) with a replayable, hash-verified audit trail.

Key properties:

• Append-only — no record is ever deleted or modified
• Rank-1 updates — memory-efficient confidence-delta tracking
• Confidence floor — configurable minimum via confidenceFloor: 0
• Replayable — the full audit trail can be replayed deterministically to reconstruct any historical state
• Hash-verified — each etch record is cryptographically linked to its predecessor

---

Provenance Metadata

Every pipeline run produces a ProvenanceMetadata bundle containing:

Field	Description
merkleRoot	Root hash of the Merkle chain for this run
entropyScore	Encoder-computed entropy/uncertainty estimate
resonanceScore	Cosine similarity score from Stigmergy recall
etchHash	Hash of the applied Holographic Etch record
provenance	Full lineage metadata (timestamps, input hash, component versions)

---

Universal Adapter Protocol (v2.1)

Adapters wire the deterministic triad to external platforms via the IMCOPAdapter contract without modifying core. The adapter pattern allows MCOP to integrate with:

• Magnific — AI image upscaling/enhancement
• Higgsfield — AI cinematic video generation
• Utopai — generative creative platform
• Generic REST/MCP/HTTP pipelines

All adapters implement the same interface:

interface IMCOPAdapter {
  encoder: NovaNeoEncoder;
  stigmergy: StigmergyV5;
  holographicEtch: HolographicEtch;
}

See docs/adapters/UNIVERSAL_ADAPTER_PROTOCOL.md for the full specification.

---

Key Design Decisions

Decision	Rationale
Cosine similarity for resonance	Scale-invariant, fast O(d), ideal for semantic vectors
Merkle chains for provenance	Tamper-evident, supports distributed verification
Rank-1 micro-etches	Memory-efficient parameter evolution tracking
Stigmergic (not direct) agent coordination	Decoupled, scalable, biologically inspired
Deterministic encoding	Reproducibility for research, debugging, and replay

---

Hardware Acceleration Seams

MCOP is designed with clear extension points for GPU/FPGA acceleration:

• Stigmergy v5 — cosine similarity search over large trace libraries benefits from GPU-accelerated matrix operations
• NOVA-NEO Encoder — high-dimensional tensor generation can leverage FPGA pipelines
• Holographic Etch — rank-1 updates are inherently parallelizable

These seams are exposed as clean interfaces in src/core/ without hard dependencies on any specific hardware backend.

---

Further Reading

• ARCHITECTURE.md — source architecture document
• docs/adr/ — Architecture Decision Records
• docs/whitepapers/ — in-depth technical whitepapers
• Core-Concepts-and-Glossary — plain-English definitions of all terms