Architecture

The technical architecture powering AI governance at scale. From trust scoring algorithms to policy enforcement pipelines.

Platform Overview

┌─────────────────────────────────────────────────────────────────────┐
│                        VORION PLATFORM                               │
├─────────────────────────────────────────────────────────────────────┤
│                                                                      │
│   USER INTERFACES                                                    │
│   ┌──────────────┐   ┌──────────────┐   ┌──────────────┐            │
│   │ AgentAnchor  │   │    Aurais    │   │    Learn     │            │
│   │  Dashboard   │   │  Assistant   │   │   Vorion     │            │
│   └──────┬───────┘   └──────┬───────┘   └──────┬───────┘            │
│          │                  │                  │                     │
│   ═══════╪══════════════════╪══════════════════╪═════════════════   │
│          │           API GATEWAY               │                     │
│   ═══════╪══════════════════╪══════════════════╪═════════════════   │
│          │                  │                  │                     │
│   ┌──────▼──────────────────▼──────────────────▼──────────────────┐ │
│   │                    COGNIGATE RUNTIME                           │ │
│   │  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐            │ │
│   │  │   Policy    │  │   Intent    │  │ Enforcement │            │ │
│   │  │  Evaluator  │  │  Processor  │  │   Engine    │            │ │
│   │  └─────────────┘  └─────────────┘  └─────────────┘            │ │
│   └────────────────────────┬───────────────────────────────────────┘ │
│                            │                                         │
│   ┌────────────────────────▼───────────────────────────────────────┐ │
│   │                     ATSF CORE RUNTIME                          │ │
│   │                                                                 │ │
│   │  ┌───────────┐  ┌───────────┐  ┌───────────┐  ┌───────────┐   │ │
│   │  │   Trust   │  │   BASIS   │  │   Proof   │  │ Governance│   │ │
│   │  │  Engine   │  │  Evaluator│  │   Plane   │  │  Engine   │   │ │
│   │  └───────────┘  └───────────┘  └───────────┘  └───────────┘   │ │
│   │                                                                 │ │
│   │  ┌───────────┐  ┌───────────┐  ┌───────────┐  ┌───────────┐   │ │
│   │  │Arbitration│  │Containment│  │ Contracts │  │ Provenance│   │ │
│   │  │  Engine   │  │  System   │  │  (Zod)    │  │  Tracker  │   │ │
│   │  └───────────┘  └───────────┘  └───────────┘  └───────────┘   │ │
│   └────────────────────────────────────────────────────────────────┘ │
│                                                                      │
│   ┌────────────────────────────────────────────────────────────────┐ │
│   │                    PERSISTENCE LAYER                            │ │
│   │  ┌───────────┐  ┌───────────┐  ┌───────────┐                   │ │
│   │  │  Memory   │  │   File    │  │  Supabase │                   │ │
│   │  │  Adapter  │  │  Adapter  │  │  Adapter  │                   │ │
│   │  └───────────┘  └───────────┘  └───────────┘                   │ │
│   └────────────────────────────────────────────────────────────────┘ │
│                                                                      │
│   ┌────────────────────────────────────────────────────────────────┐ │
│   │                     BASIS STANDARD                              │ │
│   │           6-Tier Trust Model Specification v1.0.0               │ │
│   └────────────────────────────────────────────────────────────────┘ │
│                                                                      │
└─────────────────────────────────────────────────────────────────────┘

The Three Governance Layers

Every AI action flows through three governance layers before and after execution:

1

Intent Layer

Parses agent requests, classifies action types, surfaces risks, and structures intents for policy evaluation.

  • Risk classification (LOW to CRITICAL)
  • Prompt injection detection
  • IntentRecord schema validation
2

Enforce Layer

Evaluates trust levels, applies policy rules, and makes ALLOW/DENY/ESCALATE decisions.

  • Trust tier verification
  • Policy evaluation order
  • Human escalation triggers
3

Proof Layer

Creates immutable audit trails with cryptographic hash chains for compliance and accountability.

  • SHA-256 hash chains
  • 7-10 year retention
  • Tamper-evident storage

Trust Score Calculation

Component Weights

Behavioral
40%
Compliance
25%
Identity
20%
Context
15%

Signal Processing

  • • Time-weighted averaging with 7-day half-life
  • • Recent signals weighted more heavily
  • • Last 1000 signals retained per entity
  • • Significant change threshold: 10 points

Trust Score Formula

score = round(
  behavioral × 0.40 × 1000 +
  compliance × 0.25 × 1000 +
  identity   × 0.20 × 1000 +
  context    × 0.15 × 1000
)

// Clamped to 0-1000 range
// Each component normalized to 0.0-1.0

6-Tier Trust Model

Level Name Score Range Capabilities
L0 Sandbox 0-99 Read-only, sandboxed execution
L1 Provisional 100-299 Basic operations, supervised
L2 Standard 300-499 Standard operations, external API
L3 Trusted 500-699 Cross-agent communication
L4 Certified 700-899 Admin tasks, agent spawning
L5 Autonomous 900-1000 Full autonomy, self-governance

Decay & Recovery

Trust Decay

  • • Base decay rate: 1% per interval
  • • Decay interval: 60 seconds
  • • Failure threshold: signal value < 0.3
  • • Accelerated decay: 3x after 2+ failures in 1 hour
  • • Decay formula: score × (1 - rate)^periods

Trust Recovery

  • • Success threshold: signal value ≥ 0.7
  • • Base recovery rate: 2% per success
  • • Accelerated recovery: 1.5x after 3 consecutive successes
  • • Max recovery per signal: 50 points
  • • Peak score tracking for milestone events

Deep Dive

BASIS Standard ATSF Framework Implementation Details