Arqon Anchor (v0.5.0) - Production

Status: PRODUCTION / SOTA
Latency: 77µs (Memory) / 11ms (Process)
Integrity: 100% (Zero False Positives in Benchmarks)

The Arqon Anchor is a hardware-enforced verification engine for Large Language Models. It eliminates hallucination by forcing every claim to be backed by an executable "Witness Program" that runs in a Hardware Vault before the output is released.

📚 The Library of Truth

• The Theory: Whitepaper | Runtime Substrate
• The Manuals: Direct Runtime Manual | Witness Authoring
• The Benchmarks: ArqonCore/benchmark_direct.py

🏛️ Features

Layer 1: Semantic Enforcement

• GovernanceSentinel (Sandbox): Enforces CPU and filesystem isolation.
• EvidenceGuard (Auditor): Links logical conclusions to verifiable evidence.

Layer 2: Statistical Integrity

• Adversarial Forge (Canary): Detects telepathic state leakage.
• Rarity Engine (Hypothesis E): Blocks rote memorization and disjoint reasoning via Information-Theoretic Rarity (NCD).

Layer 3: Hardware Veto

• OpaqueTunnel (Hypothesis D): Simulates hardware-locked memory to prevent optimizer peeking.
• FormalVerifier (Hypothesis G): Interface for symbolic logic proofs of state separation.

📦 Installation

pip install arqon-anchor

🚀 Usage

from anchor import run_audit

# Verify an LLM output
result = run_audit(llm_output, context="My Context", mode="omega", rarity_check=True)

if result["verified"]:
    print("✅ Truth-Locked Conclusion")
else:
    print(f"❌ Rejected: {result['error']}")

🛡️ License

MIT