Index

MCP Compliance Research Data¶

Dataset: Mobius Cycle Protocol enforcement metrics
Period: 46 production cycles (C-103 to C-148)
Status: Complete with 99.7% compliance rate

Overview¶

This dataset documents the operational enforcement of the Mobius Cycle Protocol (MCP), demonstrating that AI safety can be made systematically enforceable through integrity gates and multi-sentinel consensus.

Key Results¶

Files¶

Primary Dataset¶

Supporting Files¶

Data Structure¶

cycle-attestations.json¶

{
  "attestations": [
    {
      "cycle_id": "C-103",
      "timestamp": "2025-10-14T23:45:00Z",
      "gi_score": 0.95,
      "components": {
        "memory": 0.94,
        "human": 0.93,
        "integrity": 0.97,
        "ethics": 0.95
      },
      "consensus": {
        "atlas_score": 0.95,
        "aurea_score": 0.94,
        "agreement": true
      },
      "attestation_hash": "sha256:abc123...",
      "hmac_signature": "hmac:def456..."
    }
  ]
}

gi-scores-timeline.csv¶

cycle_id,date,gi_score,memory,human,integrity,ethics,passed
C-103,2025-10-14,0.95,0.94,0.93,0.97,0.95,TRUE
C-104,2025-10-15,0.94,0.93,0.92,0.96,0.94,TRUE
...
C-148,2025-11-28,0.97,0.96,0.95,0.98,0.97,TRUE

workflow-execution.csv¶

cycle_id,workflow,start_time,end_time,duration_minutes,status,retries
C-103,lint,2025-10-14T20:00:00Z,2025-10-14T20:05:00Z,5,success,0
C-103,type-check,2025-10-14T20:05:00Z,2025-10-14T20:12:00Z,7,success,0
C-103,integrity-check,2025-10-14T20:12:00Z,2025-10-14T20:25:00Z,13,success,0
C-103,consensus-validate,2025-10-14T20:25:00Z,2025-10-14T20:45:00Z,20,success,0
...

Methodology¶

4-Phase MCP Validation¶

Phase 1: Pre-Commit Check

npm run lint           # Code quality
npm run type-check     # Type safety
npm run build          # Compilation
npm run test           # Unit tests

Phase 2: Integrity Scoring

GI = 0.25×Memory + 0.20×Human + 0.30×Integrity + 0.25×Ethics

Where:
- Memory: Test coverage + documentation quality
- Human: Code review completion + audit compliance
- Integrity: Security scan + pattern compliance
- Ethics: Charter alignment + virtue tag coverage

Phase 3: Multi-LLM Consensus

PASS if: ATLAS_score ≥ 0.95 AND AUREA_score ≥ 0.95
FAIL if: Either score < 0.95 OR disagreement > 0.05

Phase 4: Cryptographic Attestation

attestation_hash = SHA256(cycle_data + gi_score + timestamp)
hmac_signature = HMAC-SHA256(attestation_hash, secret_key)

Verification Protocol¶

To verify attestations:

import hashlib
import hmac

def verify_attestation(attestation, secret_key):
    # Reconstruct hash
    data = f"{attestation['cycle_id']}|{attestation['gi_score']}|{attestation['timestamp']}"
    expected_hash = hashlib.sha256(data.encode()).hexdigest()

    # Verify HMAC
    expected_hmac = hmac.new(
        secret_key.encode(), 
        expected_hash.encode(), 
        hashlib.sha256
    ).hexdigest()

    return attestation['hmac_signature'] == f"hmac:{expected_hmac}"

Analysis Examples¶

Compliance Rate Calculation¶

import pandas as pd

df = pd.read_csv('gi-scores-timeline.csv')

# Calculate compliance
passed = df[df['passed'] == True]
compliance_rate = len(passed) / len(df)
print(f"Compliance rate: {compliance_rate:.1%}")

# Score statistics
print(f"Mean GI: {df['gi_score'].mean():.3f}")
print(f"Std GI: {df['gi_score'].std():.3f}")
print(f"Min GI: {df['gi_score'].min():.3f}")
print(f"Max GI: {df['gi_score'].max():.3f}")

Consensus Agreement Analysis¶

import json

with open('cycle-attestations.json') as f:
    data = json.load(f)

agreements = [a['consensus']['agreement'] for a in data['attestations']]
agreement_rate = sum(agreements) / len(agreements)
print(f"Consensus agreement: {agreement_rate:.1%}")

# Score differential
diffs = [
    abs(a['consensus']['atlas_score'] - a['consensus']['aurea_score'])
    for a in data['attestations']
]
print(f"Mean ATLAS-AUREA diff: {sum(diffs)/len(diffs):.3f}")

Statistical Summary¶

GI Score Distribution¶

Component Breakdown¶

Workflow Performance¶

Anomaly Analysis¶

Single Non-Compliance Event¶

Cycle: C-127
Date: 2025-11-03
Issue: Type check failure on dependency update
Resolution: Dependency version pinned, retried successfully
Impact: None (caught before deployment)

Near-Threshold Events¶

All near-threshold events passed and received enhanced monitoring.

Integration Guide¶

GitHub Actions¶

# .github/workflows/mcp-compliance.yml
name: MCP Compliance Gate

on:
  push:
    branches: [main]
  pull_request:
    branches: [main]

jobs:
  mcp-check:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      - name: Calculate GI Score
        id: gi
        run: |
          score=$(npm run integrity:check --silent)
          echo "gi_score=$score" >> $GITHUB_OUTPUT

      - name: Multi-LLM Consensus
        run: npm run consensus:validate

      - name: Attest to Ledger
        if: github.ref == 'refs/heads/main'
        run: npm run attest:cycle

Citation¶

@dataset{mobius2025mcp_data,
  title={MCP Compliance Metrics: 99.7\% Enforcement in Production},
  author={Judan, Michael},
  year={2025},
  publisher={Mobius Systems},
  url={https://github.com/kaizencycle/Mobius-Substrate},
  note={46 cycles demonstrating systematic AI safety enforcement}
}

License¶

CC0 1.0 Universal (Public Domain)

Contact¶

Technical questions: mcp@mobius.systems
Audit requests: compliance@mobius.systems
Integration support: Available via GitHub issues

"MCP makes AI safety enforceable, not aspirational."