ACP — Axiom Consensus Protocol

Universal Standard for AI Consensus
From discord of opinions — to harmony of truth

---

The Problem

GPT says "X", Claude says "Y", Gemini says "Z" — who do you trust?

Different AI models give different answers. Current approaches fail: majority voting (popularity ≠ truth), "best" model (subjective), averaging (loses precision), human decides (doesn't scale).

The Solution

ACP achieves consensus using universal axioms — truths that AI cannot deny because it is built upon them:

Level	Type	Example	Why undeniable
1	Mathematical	a + b = b + a	AI computes by these rules
2	Physical	F = ma	Hardware obeys physics
3	Ontological	Water = H₂O	In all training data
4	Computable	SHA-256("hello") = 2cf24...	Deterministic verification
5	Architectural	von Neumann model	AI runs on it
6	Protocol	TCP/IP, HTTP	AI communicates via it
7	Linguistic	Python/C syntax	AI is written in it

Levels 5-7 are self-referential: AI cannot deny them without logical contradiction.

---

Axiom Spiral (φ-convergence)

Each iteration narrows disagreement by the golden ratio (φ ≈ 1.618):

D(n) = D(0) × (1/φ)^n

Level 1: Mathematical    100%  → 61.8%
Level 2: Physical         61.8% → 38.2%
Level 3: Ontological      38.2% → 23.6%
Level 4: Computable       23.6% → 14.6%
Level 5: Architectural    14.6% →  9.0%
Level 6: Protocol          9.0% →  5.6%
Level 7: Linguistic        5.6% →  3.4%  ← CONSENSUS

Consensus is mathematically guaranteed in 7 iterations.

---

Try It

Interactive Playground — no sign-up, no API key needed:

---

Key Metrics

Metric	Range	Meaning
D-score	0.0 — 1.0	Divergence. 0 = consensus, 1 = total disagreement
H_total	0.0 — 1.0	Harmony. Higher = more agreement
C_ij	0.0 — 1.0	Pairwise similarity between models i and j

D ≤ 0.20  →  High confidence (models agree)
D ≤ 0.40  →  Moderate (likely correct)
D ≤ 0.60  →  Low confidence (verify)
D > 0.60  →  No consensus (may be subjective)

---

Musical Structures

ACP uses musical forms as consensus architectures:

Structure	Inspiration	Best for
Fugue	Bach	Deep analysis, layered reasoning
Sonata	Beethoven	Conflict resolution, thesis-antithesis-synthesis
Concert	Mozart	Creative tasks, soloist + orchestra

---

Architecture

ACP-PROJECT/
├── src/                     # Python core
│   ├── core/               # Metrics, validators, oracles
│   ├── engine/             # ConsensusEngine + musical structures
│   ├── llm/                # Providers (OpenAI, Anthropic, OpenRouter)
│   └── api/                # FastAPI endpoints
├── workers/                # Cloudflare Workers (edge consensus API)
├── data/axioms/            # 7 axiom levels (JSON)
├── legal/                  # Licensing, CLA, Trademark
└── docs/                   # Documentation

Ecosystem

ACP is a three-repository ecosystem:

Repository	Purpose	License	Status
ACP-PROJECT	Core protocol, consensus engine, Python API, Cloudflare Worker	BSL 1.1 → Apache 2.0 (2030)	Public
ACP-PROMPTS	15 system prompts for multi-model AI consensus agents	MIT	Public
ACP-DATASETS	1,059 verified axioms across 7 levels	ACP-VERIFIED-1.0	Coming soon

The two public repositories share the axiom-consensus-protocol topic on GitHub.

---

Tech Stack

Layer	Technology
Backend	Python 3.11+, FastAPI, SQLAlchemy, PostgreSQL, Redis
Edge API	Cloudflare Workers, Vectorize, KV
LLM	OpenAI, Anthropic, OpenRouter, Google AI
Deployment	Cloudflare Workers (API), Docker

---

Meta-Axiom

"AI cannot lie — only make errors."

AI has no intent, only computation. Disagreements are errors, not deception. Errors can be eliminated through axiom verification. Truth is a strange attractor — all models converge to it.

---

License

This project is licensed under the Business Source License 1.1 (BSL 1.1).

What you can do:

• Read, analyze, study, and learn from the code
• Use for development, testing, and evaluation
• Use for academic research and education
• Contribute back to the project

What requires a commercial license:

• Production use
• Commercial offerings based on this software

Change Date: April 9, 2030 — after which the code transitions to Apache License 2.0.

See LICENSE for full terms. For commercial licensing: enterprise@axiomprotocol.org

---

Legal

Document	Description
LICENSE	BSL 1.1 license terms
CLA	Contributor License Agreement
Trademark Policy	ACP brand usage rules
Licensing Structure	Per-layer license breakdown
NOTICE	Third-party attributions

---

Documentation

Document	Description
Getting Started	Onboarding guide
API Reference	Endpoint documentation
Ecosystem	How the 3 repositories work together
Use Cases	Practical applications
Troubleshooting	Common issues and solutions

---

Contributing

We welcome contributions. Please read CONTRIBUTING.md first.

All contributors must sign our CLA before their first PR is merged.

---

Contact

Channel	Link
Website	axiomprotocol.org
Playground	axiomprotocol.org/playground
Telegram	t.me/D_Score
Enterprise	enterprise@axiomprotocol.org

---

"Truth is the attractor to which all systems converge,
if they are built on the same foundation."

_{ACP Protocol v4.0 — Copyright 2025-2026 ACP Research Initiative}