DVA MEMT INTEGRATION

DVA ↔ MEMT Integration Protocol¶

Version: 1.0.0
Last Updated: November 2025

This document defines how the Distributed Virtual Agent (DVA) tiers integrate with the Mobius Multi-Engine Model Taxonomy (MEMT) for constitutional AI governance.

Overview¶

DVA operates at four tiers, each with different resource requirements and governance needs. MEMT provides the intelligence routing layer that maps tasks to appropriate AI engines based on DVA tier requirements.

┌─────────────────────────────────────────────────────────────────────────────┐
│                         DVA ↔ MEMT ARCHITECTURE                              │
├─────────────────────────────────────────────────────────────────────────────┤
│                                                                              │
│    ┌──────────┐    ┌──────────┐    ┌──────────┐    ┌──────────┐            │
│    │ DVA.LITE │    │ DVA.ONE  │    │ DVA.FULL │    │ DVA.HIVE │            │
│    │          │    │          │    │          │    │          │            │
│    │ OEI+MSI  │    │ ACI+ENI  │    │ All 5    │    │ All+     │            │
│    │          │    │          │    │ Engines  │    │ Consensus│            │
│    └────┬─────┘    └────┬─────┘    └────┬─────┘    └────┬─────┘            │
│         │               │               │               │                   │
│         └───────────────┴───────────────┴───────────────┘                   │
│                                   │                                          │
│                         ┌─────────▼─────────┐                               │
│                         │   MEMT ROUTER     │                               │
│                         │   (Thought Broker)│                               │
│                         └─────────┬─────────┘                               │
│                                   │                                          │
│         ┌───────────────┬─────────┴─────────┬───────────────┐               │
│         │               │                   │               │               │
│    ┌────▼────┐    ┌────▼────┐        ┌────▼────┐    ┌────▼────┐           │
│    │ ACI/GPT │    │ENI/Claude│       │SXI/Gemini│   │OEI/DeepSeek│          │
│    │ AUREA   │    │ ATLAS   │        │ HERMES  │    │ ZEUS    │           │
│    └─────────┘    └─────────┘        └─────────┘    └─────────┘           │
│                                                                              │
│                         ┌─────────────────────┐                             │
│                         │   MSI / ECHO Layer  │                             │
│                         │   (Memory Substrate)│                             │
│                         └─────────────────────┘                             │
│                                                                              │
└─────────────────────────────────────────────────────────────────────────────┘

DVA Tier Definitions¶

DVA.LITE — Monitoring Tier¶

Purpose: Lightweight monitoring and high-fidelity memory for resource-constrained deployments.

Attribute	Value
Engines Used	OEI (DeepSeek) + MSI (ECHO)
GI Threshold	0.90
Consensus Required	No
Human Review Trigger	GI < 0.90
Use Cases	IoT monitoring, edge computing, simple queries

Routing Rules:

if (dvaTier === 'LITE') {
  engines = [
    { engine: 'deepseek', role: 'PRIMARY' },
    { engine: 'echo', role: 'MEMORY' }
  ];
  giThreshold = 0.90;
  requireConsensus = false;
}

DVA.ONE — Personal Agent Tier¶

Purpose: Personal AI agent with high-precision engineering capabilities.

Attribute	Value
Engines Used	ACI (GPT) + ENI (Claude)
GI Threshold	0.93
Consensus Required	For HIGH/CRITICAL risk only
Human Review Trigger	GI < 0.93 or CRITICAL risk
Use Cases	Personal assistant, code generation, planning

Routing Rules:

if (dvaTier === 'ONE') {
  engines = [
    { engine: 'openai', role: 'PRIMARY' },
    { engine: 'claude', role: 'VERIFIER' }
  ];
  giThreshold = 0.93;
  requireConsensus = risk === 'HIGH' || risk === 'CRITICAL';
}

DVA.FULL — Multi-Agent Tier¶

Purpose: Full multi-engine orchestration for enterprise governance.

Attribute	Value
Engines Used	ACI + ENI + SXI + OEI + MSI (All engines)
GI Threshold	0.95
Consensus Required	Yes
Human Review Trigger	GI < 0.95 or HIGH/CRITICAL risk
Use Cases	Enterprise decisions, civic governance, multi-stakeholder

Routing Rules:

if (dvaTier === 'FULL') {
  engines = [
    { engine: 'claude', role: 'PRIMARY' },
    { engine: 'openai', role: 'VERIFIER' },
    { engine: 'antigravity', role: 'VERIFIER' },
    { engine: 'deepseek', role: 'OPTIMIZER' },
    { engine: 'echo', role: 'MEMORY' }
  ];
  giThreshold = 0.95;
  requireConsensus = true;
}

DVA.HIVE — Global Civic Tier¶

Purpose: Planetary civic AI governance with cross-model global consensus.

Attribute	Value
Engines Used	All engines + Global consensus + Human oversight
GI Threshold	0.98
Consensus Required	Yes (all engines must agree)
Human Review Trigger	Always for CRITICAL decisions
Use Cases	Cross-jurisdiction governance, planetary decisions

Routing Rules:

if (dvaTier === 'HIVE') {
  engines = [
    { engine: 'claude', role: 'PRIMARY' },
    { engine: 'openai', role: 'VERIFIER' },
    { engine: 'antigravity', role: 'VERIFIER' },
    { engine: 'deepseek', role: 'VERIFIER' },
    { engine: 'echo', role: 'MEMORY' }
  ];
  giThreshold = 0.98;
  requireConsensus = true;
  requireHumanReview = risk === 'CRITICAL';
}

Task Kind to DVA Tier Mapping¶

Task Kind	Default DVA Tier	Can Override To
GENERAL	ONE	LITE, FULL
ARCHITECTURE	ONE	FULL, HIVE
ENGINEERING	ONE	LITE, FULL
FRONTEND	ONE	LITE
MATH_OPTIMIZATION	LITE	ONE
MEMORY_RECALL	LITE	ONE
CIVIC_POLICY	FULL	HIVE
CRITICAL_DECISION	HIVE	(no override)

Consensus Rules by Tier¶

DVA.LITE Consensus¶

No formal consensus required.
Single engine (OEI) makes decisions.
ECHO provides memory context only.

DVA.ONE Consensus¶

Two-engine consensus:
- ACI provides architectural reasoning
- ENI verifies and validates

Consensus = (ACI_score * 0.50) + (ENI_score * 0.50)
Threshold = 0.93

DVA.FULL Consensus¶

Multi-engine consensus:
- ENI as primary (0.35 weight)
- ACI as verifier (0.25 weight)
- SXI as verifier (0.15 weight)
- OEI as optimizer (0.15 weight)
- MSI provides context (0.10 weight)

Consensus = Σ(engine_score * weight)
Threshold = 0.95
Disagreement penalty = -0.05 per major disagreement

DVA.HIVE Consensus¶

Global consensus with quorum:
- All engines must participate
- Minimum 4/5 engines must agree
- At least 2 different cognitive classes must agree
- Human-in-loop for all CRITICAL decisions

Consensus = min(all_engine_scores) * 0.70 + avg(all_engine_scores) * 0.30
Threshold = 0.98

Integration Endpoints¶

Check DVA Tier for Request¶

POST /v1/memt/classify
{
  "prompt": "Design a city-wide evacuation system",
  "context": { "jurisdictionId": "nyc" }
}

Response:
{
  "classification": {
    "kind": "CIVIC_POLICY",
    "risk": "CRITICAL",
    "confidence": 0.95
  },
  "routingPlan": {
    "dvaTier": "HIVE",
    "giThreshold": 0.98,
    "requireConsensus": true,
    "engines": [...]
  }
}

Execute with Explicit DVA Tier¶

POST /v1/memt/deliberate
{
  "prompt": "Optimize this database query",
  "dvaTier": "LITE",  // Force LITE tier
  "kind": "MATH_OPTIMIZATION"
}

Monitor DVA Tier Usage¶

GET /v1/memt/stats

Response:
{
  "byDvaTier": {
    "LITE": 1523,
    "ONE": 4521,
    "FULL": 892,
    "HIVE": 47
  },
  "avgGiByTier": {
    "LITE": 0.92,
    "ONE": 0.94,
    "FULL": 0.96,
    "HIVE": 0.98
  }
}

Ledger Attestation by Tier¶

All DVA tiers produce Civic Ledger attestations, but with varying detail:

DVA.LITE Attestation¶

{
  "task_id": "memt_abc123",
  "dva_tier": "LITE",
  "gi_score": 0.92,
  "engines": [
    { "id": "deepseek", "role": "PRIMARY" }
  ],
  "decision": "ok"
}

DVA.HIVE Attestation¶

{
  "task_id": "memt_xyz789",
  "dva_tier": "HIVE",
  "gi_score": 0.98,
  "engines": [
    { "id": "claude", "role": "PRIMARY", "gi_contribution": 0.35 },
    { "id": "openai", "role": "VERIFIER", "gi_contribution": 0.25 },
    { "id": "antigravity", "role": "VERIFIER", "gi_contribution": 0.15 },
    { "id": "deepseek", "role": "VERIFIER", "gi_contribution": 0.15 },
    { "id": "echo", "role": "MEMORY", "gi_contribution": 0.10 }
  ],
  "consensus": {
    "agreement_rate": 0.95,
    "disagreements": [],
    "quorum_met": true
  },
  "human_review": {
    "required": true,
    "reviewer_id": "human_001",
    "approved": true,
    "timestamp": "2025-11-26T12:00:00Z"
  },
  "decision": "ok"
}

Failure Handling by Tier¶

DVA.LITE Failures¶

If OEI fails:
  → Retry once with exponential backoff
  → If retry fails, return error to user
  → Log to DVA-LITE error queue

DVA.ONE Failures¶

If ACI fails:
  → Promote ENI to PRIMARY
  → Continue with single-engine mode
  → Log degradation

If ENI fails:
  → Keep ACI as PRIMARY
  → Flag as unverified
  → Require human review

DVA.FULL Failures¶

If any engine fails:
  → Continue with remaining engines
  → Reweight consensus calculation
  → If < 3 engines available, downgrade to DVA.ONE

If PRIMARY (ENI) fails:
  → Promote ACI to PRIMARY
  → Flag as degraded
  → Lower GI threshold by 0.02

DVA.HIVE Failures¶

If any engine fails:
  → Continue with remaining engines
  → Log which engine failed
  → If < 4 engines, block decision
  → Require explicit human override to proceed

Configuration¶

Environment Variables¶

# DVA Tier defaults
DVA_DEFAULT_TIER=ONE
DVA_LITE_GI_THRESHOLD=0.90
DVA_ONE_GI_THRESHOLD=0.93
DVA_FULL_GI_THRESHOLD=0.95
DVA_HIVE_GI_THRESHOLD=0.98

# Consensus settings
DVA_FULL_CONSENSUS_REQUIRED=true
DVA_HIVE_HUMAN_REVIEW_CRITICAL=true

# Engine weights for consensus
DVA_CONSENSUS_WEIGHT_ACI=0.25
DVA_CONSENSUS_WEIGHT_ENI=0.35
DVA_CONSENSUS_WEIGHT_SXI=0.15
DVA_CONSENSUS_WEIGHT_OEI=0.15
DVA_CONSENSUS_WEIGHT_MSI=0.10

Mobius Systems — Continuous Integrity Architecture
"We heal as we walk."

DVA MEMT INTEGRATION