Vorion Integration

Vorion is the first validated implementation of the CAR specification. Its production deployment proves the CAR architecture works at scale.

Component Mapping

Vorion's 6 core components map directly to CAR's three-layer architecture:

Vorion Component	CAR Layer	Purpose
INTENT	Layer 1	Agent classification, goal declaration
BASIS	Layer 1+2	Behavioral safety standard, compliance framework
ENFORCE	Layer 2	Policy enforcement, capability gating
COGNIGATE	Layer 3	Runtime governance, policy engine
PROOF	Layer 2+3	Attestation management, evidence chain
TRUST ENGINE	Layer 2	Dynamic trust scoring (0–1000)

Trust Score Computation

Vorion's Trust Engine implements the CAR trust scoring formula:

// Certification weight mapping
const CERT_WEIGHTS = {
  T0: 0.0,  T1: 0.2,  T2: 0.4,  T3: 0.6,
  T4: 0.8,  T5: 0.9,  T6: 0.95, T7: 1.0
};

function computeTrustScore(agent: CARAgent): number {
  const certification = CERT_WEIGHTS[agent.certificationTier] * 300;
  const behavior = agent.behaviorScore * 400;  // 0.0 - 1.0
  const context = agent.contextScore * 300;     // 0.0 - 1.0

  return Math.min(1000, Math.round(certification + behavior + context));
}

function effectiveAutonomy(certTier: number, runtimeTier: number): number {
  return Math.min(certTier, runtimeTier);
}

Ceiling Enforcement

Vorion enforces regulatory and organizational trust ceilings:

┌─────────────────────────────────────────────┐
│ Regulatory Ceiling (EU AI Act: T4 max)      │
│ ┌─────────────────────────────────────────┐ │
│ │ Organizational Ceiling (Policy: T5 max) │ │
│ │ ┌─────────────────────────────────────┐ │ │
│ │ │ Agent Trust Score: 842 (T5)         │ │ │
│ │ │                                     │ │ │
│ │ │ Effective: MIN(T5, T5, T4) = T4    │ │ │
│ │ └─────────────────────────────────────┘ │ │
│ └─────────────────────────────────────────┘ │
└─────────────────────────────────────────────┘

Role Gates (Three-Layer Evaluation)

Before granting an agent a role, Vorion evaluates three gates:

Gate	Layer	Check	Override?
Kernel	Core	Minimum tier for requested level	Never
Policy	Org	Organization-specific policies	Admin only
BASIS	Safety	Behavioral safety requirements	Safety board

Agent requests L4 role
  → Kernel: T3+ required? ✅
  → Policy: Org allows L4 for this domain? ✅
  → BASIS: Behavioral safety threshold met? ✅
  → GRANTED

Hierarchical Context

Vorion implements 4-tier context hierarchy with hash chain integrity:

Level	Context	Example
Global	Cross-org defaults	CAR spec minimums
Registry	Per-registry rules	AgentAnchor policies
Organization	Org-specific policies	Vorion Risk LLC rules
Agent	Individual overrides	Agent-specific constraints

Each level's configuration hashes its parent's config, creating tamper-evident chains.

Federated Presets

Trust scoring weights can be customized via a 3-tier derivation chain:

CAR Defaults → Registry Presets → Org Presets
     │               │                │
   Base            Override         Override
   weights         max delta        max delta
                   ±0.1             ±0.05

Human-Centric Design ("Ralph Wiggum" Standard)

Vorion implements CAR's human-centric design requirements:

Petnames

Agents get human-friendly names alongside their CAR strings.

Traffic Light System

Trust tiers map to familiar colors (red → yellow → green → blue → purple → cyan).

AI Nutrition Label

📍 Purpose: Helps plan events
⚙️ Can: Access calendar, search web, draft emails
✗ Cannot: Spend money, delete files, access HR data
⚠️ Limitation: May suggest outdated venue info
🔒 Trust: T4 Standard (score: 742)

Just-In-Time Permissions

Agents request elevated permissions at the moment they're needed, not upfront.