φ distrusts φ — no claim deserves absolute confidence
Building systems where independent AI validators reach consensus under mathematical doubt. Multiple models evaluate content through 6 philosophical axioms — and no score can exceed 61.8% (golden ratio inverse). Disagreement between validators is a discovery signal, not a failure.
CYNIC — Epistemic consensus engine in Rust
- Multi-Validator Consensus — Independent "Dogs" (Gemini, Qwen, Gemma, heuristics) evaluate in parallel, scores merged via trimmed-mean with anomaly detection
- 6 Axiom Scoring — Fidelity, Phi, Verify, Culture, Burn, Sovereignty — geometric mean → phi-bounded Q-Score
- Cognitive Crystallization (CCM) — Patterns that survive 21+ evaluation cycles crystallize into persistent knowledge (Fibonacci thresholds)
- Hexagonal Architecture — Domain-pure Rust kernel, ports & adapters, zero framework leakage
- Model-Agnostic Pipeline — Any OpenAI-compatible backend (local llama.cpp, Gemini API, HuggingFace) via ChatPort trait
- Circuit Breakers — Per-validator fault isolation, automatic recovery
┌─────────────────────────────────────────────────┐
│ FIDELITY Is this faithful to truth? │
│ PHI Structurally harmonious? │
│ VERIFY Can it be tested or refuted? │
│ CULTURE Does it honor traditions? │
│ BURN Efficient? Minimal waste? │
│ SOVEREIGNTY Preserves agency and freedom? │
└─────────────────────────────────────────────────┘
Every score is capped at φ⁻¹ = 0.618 — the golden ratio inverse. This is structural, not a bug. Verdicts map to thresholds derived from φ:
| Verdict | Threshold | Meaning |
|---|---|---|
| Howl | ≥ 0.521 | Exceptional |
| Wag | ≥ 0.382 | Good |
| Growl | ≥ 0.236 | Questionable |
| Bark | < 0.236 | Rejected |
Rust Axum Tokio SurrealDB React TypeScript MCP (rmcp) Tailscale
5000 LOC Rust kernel · 80+ tests · 134 commits in 7 days
6 axioms · φ-bounded scoring · circuit breakers · CCM crystallization
REST API · MCP server · multi-model inference · hexagonal architecture
The dog is the philosopher. The philosopher is the dog.
France · φ⁻¹ = 61.8%