rust-kgdb 0.5.3 → 0.5.5

package/examples/underwriting-agent.js

@@ -1,13 +1,16 @@
 /**
- * ============================================================================
  * UNDERWRITING AGENT - Built on HyperMind Framework
- * ============================================================================
  *
- * This agent demonstrates the HyperMind neuro-symbolic architecture:
+ * This example demonstrates proper agentic flow using the HyperMind framework:
  *
- *   TYPE THEORY      →  All tools have typed signatures (A → B)
- *   CATEGORY THEORY  →  Tools compose as morphisms (f ∘ g)
- *   PROOF THEORY     →  Every execution produces an audit witness
+ *   ┌─────────────────────────────────────────────────────────────────────┐
+ *   │                    HyperMind Framework Architecture                 │
+ *   │                                                                     │
+ *   │   TYPE THEORY      →  Tools have typed signatures (A → B)          │
+ *   │   CATEGORY THEORY  →  Tools compose as morphisms (f ∘ g)           │
+ *   │   PROOF THEORY     →  Every execution produces audit witness       │
+ *   │                                                                     │
+ *   └─────────────────────────────────────────────────────────────────────┘
  *
  * Real-World Data Sources:
  *   - ISO (Insurance Services Office) rating factors
@@ -15,534 +18,559 @@
  *   - US Census Bureau NAICS industry codes
  *   - FEMA catastrophe exposure zones
  *
- * Industry Standards Applied:
- *   - Loss ratio threshold: 0.70 (industry standard referral trigger)
- *   - Experience modification: 5/10 year breaks
- *   - Territory factors: Catastrophe zones (hurricane, earthquake, tornado)
- *   - Premium formula: ISO standard methodology
+ * Requirements:
+ *   - Set ANTHROPIC_API_KEY or OPENAI_API_KEY environment variable
+ *   - Or use model: 'mock' for offline testing
  *
- * ============================================================================
+ * Usage:
+ *   ANTHROPIC_API_KEY=sk-... node examples/underwriting-agent.js
+ *   OPENAI_API_KEY=sk-... node examples/underwriting-agent.js --model gpt-4o
+ *   node examples/underwriting-agent.js --model mock
  */
 
+const { HyperMindAgent } = require('../hypermind-agent.js')
 const {
   GraphDB,
-  GraphFrame,
   EmbeddingService,
   DatalogProgram,
   evaluateDatalog,
+  GraphFrame,
   getVersion
 } = require('../index.js')
 
-// ============================================================================
-// REAL-WORLD COMMERCIAL INSURANCE DATA
-// Based on ISO rating factors and NAIC guidelines
-// ============================================================================
+// ═══════════════════════════════════════════════════════════════════════════
+// CONFIGURATION
+// ═══════════════════════════════════════════════════════════════════════════
+
+const MODEL = process.argv.includes('--model')
+  ? process.argv[process.argv.indexOf('--model') + 1]
+  : process.env.ANTHROPIC_API_KEY ? 'claude-sonnet-4'
+  : process.env.OPENAI_API_KEY ? 'gpt-4o'
+  : 'mock'
 
-const UNDERWRITING_DATA_TTL = `
-@prefix : <http://underwriting.org/> .
+// ═══════════════════════════════════════════════════════════════════════════
+// UNDERWRITING KNOWLEDGE BASE (ISO/NAIC-Informed Data)
+// ═══════════════════════════════════════════════════════════════════════════
+
+const UNDERWRITING_KB = `
+@prefix uw: <http://underwriting.org/> .
+@prefix naics: <http://naics.com/code/> .
+@prefix fema: <http://fema.gov/zone/> .
 @prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
 @prefix xsd: <http://www.w3.org/2001/XMLSchema#> .
-@prefix iso: <http://iso.com/rating/> .
-@prefix naics: <http://census.gov/naics/> .
-
-# === BUSINESS ENTITIES (Real NAICS codes from US Census Bureau) ===
-
-:BUS001 rdf:type :CommercialAccount ;
-        :name "Acme Manufacturing Inc" ;
-        :naicsCode "332119" ;
-        :naicsDescription "Metal Stamping" ;
-        :employees "250"^^xsd:integer ;
-        :annualRevenue "45000000"^^xsd:decimal ;
-        :territory "FL-33101" ;
-        :yearsInBusiness "15"^^xsd:integer ;
-        :safetyProgram "true"^^xsd:boolean .
-
-:BUS002 rdf:type :CommercialAccount ;
-        :name "TechStart LLC" ;
-        :naicsCode "541512" ;
-        :naicsDescription "Computer Systems Design" ;
-        :employees "45"^^xsd:integer ;
-        :annualRevenue "8500000"^^xsd:decimal ;
-        :territory "CA-94105" ;
-        :yearsInBusiness "3"^^xsd:integer ;
-        :cyberCoverage "required"^^xsd:boolean .
-
-:BUS003 rdf:type :CommercialAccount ;
-        :name "SafeHaul Logistics" ;
-        :naicsCode "484121" ;
-        :naicsDescription "General Freight Trucking, Long-Distance" ;
-        :employees "120"^^xsd:integer ;
-        :annualRevenue "22000000"^^xsd:decimal ;
-        :territory "TX-75201" ;
-        :yearsInBusiness "8"^^xsd:integer ;
-        :fleetSize "45"^^xsd:integer .
-
-:BUS004 rdf:type :CommercialAccount ;
-        :name "Downtown Restaurant Group" ;
-        :naicsCode "722511" ;
-        :naicsDescription "Full-Service Restaurants" ;
-        :employees "85"^^xsd:integer ;
-        :annualRevenue "12000000"^^xsd:decimal ;
-        :territory "NY-10001" ;
-        :yearsInBusiness "12"^^xsd:integer ;
-        :liquorLicense "true"^^xsd:boolean .
-
-# === ISO RATING FACTORS (Industry standard base rates) ===
-
-iso:NAICS-332119 rdf:type :RatingClass ;
-                 :baseRate "2.45"^^xsd:decimal ;
-                 :modificationClass "manufacturing" ;
-                 :hazardGrade "B" ;
-                 :description "Metal stamping - moderate machinery risk" .
-
-iso:NAICS-541512 rdf:type :RatingClass ;
-                 :baseRate "0.85"^^xsd:decimal ;
-                 :modificationClass "technology" ;
-                 :hazardGrade "A" ;
-                 :description "Professional services - low physical risk" .
-
-iso:NAICS-484121 rdf:type :RatingClass ;
-                 :baseRate "3.80"^^xsd:decimal ;
-                 :modificationClass "transportation" ;
-                 :hazardGrade "C" ;
-                 :description "Trucking - high accident frequency" .
-
-iso:NAICS-722511 rdf:type :RatingClass ;
-                 :baseRate "1.95"^^xsd:decimal ;
-                 :modificationClass "hospitality" ;
-                 :hazardGrade "B" ;
-                 :description "Restaurant - slip/fall, food safety" .
-
-# === TERRITORY FACTORS (FEMA catastrophe zones) ===
-
-:TERR-FL-33101 rdf:type :TerritoryRating ;
-               :territoryMod "1.35"^^xsd:decimal ;
-               :catExposure "hurricane" ;
-               :femaZone "V" ;
-               :description "Miami-Dade - high hurricane exposure" .
-
-:TERR-CA-94105 rdf:type :TerritoryRating ;
-               :territoryMod "1.15"^^xsd:decimal ;
-               :catExposure "earthquake" ;
-               :femaZone "Seismic-4" ;
-               :description "San Francisco - earthquake zone" .
-
-:TERR-TX-75201 rdf:type :TerritoryRating ;
-               :territoryMod "1.05"^^xsd:decimal ;
-               :catExposure "tornado" ;
-               :femaZone "Tornado-Alley" ;
-               :description "Dallas - moderate tornado exposure" .
-
-:TERR-NY-10001 rdf:type :TerritoryRating ;
-               :territoryMod "1.25"^^xsd:decimal ;
-               :catExposure "flood" ;
-               :femaZone "AE" ;
-               :description "Manhattan - flood zone, high density" .
-
-# === LOSS HISTORY (5-year experience) ===
-
-:BUS001 :lossCount "2"^^xsd:integer ;
-        :totalLossAmount "125000"^^xsd:decimal ;
-        :lossRatio "0.45"^^xsd:decimal ;
-        :largestClaim "85000"^^xsd:decimal .
-
-:BUS002 :lossCount "0"^^xsd:integer ;
-        :totalLossAmount "0"^^xsd:decimal ;
-        :lossRatio "0.00"^^xsd:decimal ;
-        :largestClaim "0"^^xsd:decimal .
-
-:BUS003 :lossCount "4"^^xsd:integer ;
-        :totalLossAmount "380000"^^xsd:decimal ;
-        :lossRatio "0.72"^^xsd:decimal ;
-        :largestClaim "145000"^^xsd:decimal .
-
-:BUS004 :lossCount "1"^^xsd:integer ;
-        :totalLossAmount "45000"^^xsd:decimal ;
-        :lossRatio "0.28"^^xsd:decimal ;
-        :largestClaim "45000"^^xsd:decimal .
-`
-
-// ============================================================================
-// HYPERMIND TOOL: Risk Embedding Generator (Morphism: RiskProfile → Vector)
-// Type signature enforced at compile-time
-// ============================================================================
 
-function generateRiskEmbedding(industryClass, lossRatio, territoryMod, yearsInBusiness) {
-  const embedding = new Array(384).fill(0)
+# Business Accounts with NAICS Industry Codes
+uw:BUS001 rdf:type uw:BusinessAccount ;
+          uw:name "Acme Manufacturing Inc." ;
+          uw:naicsCode naics:332710 ;
+          uw:industry "Machine Shops" ;
+          uw:yearsInBusiness "15"^^xsd:integer ;
+          uw:annualRevenue "8500000"^^xsd:decimal ;
+          uw:employeeCount "45"^^xsd:integer ;
+          uw:lossRatio "0.45"^^xsd:float ;
+          uw:femaZone fema:X ;
+          uw:territoryCode "FL-027" .
+
+uw:BUS002 rdf:type uw:BusinessAccount ;
+          uw:name "TechStart LLC" ;
+          uw:naicsCode naics:541511 ;
+          uw:industry "Custom Computer Programming" ;
+          uw:yearsInBusiness "3"^^xsd:integer ;
+          uw:annualRevenue "1200000"^^xsd:decimal ;
+          uw:employeeCount "12"^^xsd:integer ;
+          uw:lossRatio "0.15"^^xsd:float ;
+          uw:femaZone fema:C ;
+          uw:territoryCode "CA-037" .
+
+uw:BUS003 rdf:type uw:BusinessAccount ;
+          uw:name "SafeHaul Logistics" ;
+          uw:naicsCode naics:484110 ;
+          uw:industry "General Freight Trucking" ;
+          uw:yearsInBusiness "8"^^xsd:integer ;
+          uw:annualRevenue "4200000"^^xsd:decimal ;
+          uw:employeeCount "28"^^xsd:integer ;
+          uw:lossRatio "0.72"^^xsd:float ;
+          uw:femaZone fema:AE ;
+          uw:territoryCode "TX-201" .
+
+uw:BUS004 rdf:type uw:BusinessAccount ;
+          uw:name "Downtown Restaurant Group" ;
+          uw:naicsCode naics:722511 ;
+          uw:industry "Full-Service Restaurants" ;
+          uw:yearsInBusiness "12"^^xsd:integer ;
+          uw:annualRevenue "3100000"^^xsd:decimal ;
+          uw:employeeCount "65"^^xsd:integer ;
+          uw:lossRatio "0.28"^^xsd:float ;
+          uw:femaZone fema:X500 ;
+          uw:territoryCode "NY-061" .
+
+# ISO Base Rates by Industry Class
+naics:332710 uw:isoBaseRate "12.50"^^xsd:decimal ;
+             uw:hazardClass "Manufacturing" ;
+             uw:riskTier "Medium" .
+
+naics:541511 uw:isoBaseRate "4.25"^^xsd:decimal ;
+             uw:hazardClass "Office/Professional" ;
+             uw:riskTier "Low" .
+
+naics:484110 uw:isoBaseRate "18.75"^^xsd:decimal ;
+             uw:hazardClass "Transportation" ;
+             uw:riskTier "High" .
+
+naics:722511 uw:isoBaseRate "8.90"^^xsd:decimal ;
+             uw:hazardClass "Restaurant/Hospitality" ;
+             uw:riskTier "Medium" .
+
+# FEMA Zone Risk Multipliers
+fema:X uw:floodRisk "Minimal" ;
+       uw:territoryMultiplier "1.0"^^xsd:float .
+
+fema:C uw:floodRisk "Moderate" ;
+       uw:territoryMultiplier "1.15"^^xsd:float .
+
+fema:AE uw:floodRisk "High - 1% Annual Chance" ;
+        uw:territoryMultiplier "1.45"^^xsd:float .
+
+fema:X500 uw:floodRisk "Moderate - 0.2% Annual Chance" ;
+          uw:territoryMultiplier "1.10"^^xsd:float .
+
+# Territory Modifiers (based on state/region)
+uw:Territory_FL uw:hurricaneExposure "High" ;
+                uw:modificationFactor "1.35"^^xsd:float .
+
+uw:Territory_CA uw:earthquakeExposure "High" ;
+                uw:modificationFactor "1.25"^^xsd:float .
+
+uw:Territory_TX uw:hurricaneExposure "Moderate" ;
+                uw:modificationFactor "1.20"^^xsd:float .
+
+uw:Territory_NY uw:litigationRisk "High" ;
+                uw:modificationFactor "1.30"^^xsd:float .
+`
 
-  // Industry risk encoding (dims 0-95) - based on ISO hazard grades
-  const industryRiskFactors = {
-    manufacturing: 0.60,   // Moderate - machinery hazards
-    technology: 0.25,      // Low - office environment
-    transportation: 0.85,  // High - vehicle accidents
-    hospitality: 0.55      // Moderate - public liability
-  }
-  const industryRisk = industryRiskFactors[industryClass] || 0.5
-  for (let i = 0; i < 96; i++) {
-    embedding[i] = industryRisk * (0.7 + Math.sin(i * 0.1) * 0.3)
-  }
+// ═══════════════════════════════════════════════════════════════════════════
+// TOOL IMPLEMENTATIONS (Typed Morphisms)
+// ═══════════════════════════════════════════════════════════════════════════
 
-  // Loss ratio encoding (dims 96-191) - key underwriting metric
-  for (let i = 96; i < 192; i++) {
-    embedding[i] = lossRatio * (0.6 + Math.cos((i - 96) * 0.08) * 0.4)
-  }
+/**
+ * Tool: kg.sparql.query
+ * Type: SPARQLQuery → BindingSet
+ */
+async function sparqlQuery(db, query) {
+  const results = db.querySelect(query)
+  return results.map(r => r.bindings)
+}
 
-  // Territory risk encoding (dims 192-287)
-  const normalizedTerritory = (territoryMod - 1.0) / 0.35  // Normalize 1.0-1.35 to 0-1
-  for (let i = 192; i < 288; i++) {
-    embedding[i] = normalizedTerritory * (0.5 + Math.sin((i - 192) * 0.12) * 0.5)
+/**
+ * Tool: kg.graphframe.analyze
+ * Type: Graph → AnalysisResult
+ */
+function graphframeAnalyze(vertices, edges) {
+  // GraphFrame constructor requires JSON strings
+  const gf = new GraphFrame(JSON.stringify(vertices), JSON.stringify(edges))
+  // pageRank and connectedComponents return JSON strings
+  const pageRankResult = JSON.parse(gf.pageRank(0.85, 20))
+  const componentsResult = JSON.parse(gf.connectedComponents())
+  return {
+    pageRank: pageRankResult.ranks || pageRankResult,
+    components: componentsResult.components || componentsResult
   }
+}
 
-  // Experience factor (dims 288-383) - more years = lower risk
-  const experienceFactor = Math.min(yearsInBusiness / 20, 1)  // Cap at 20 years
-  for (let i = 288; i < 384; i++) {
-    embedding[i] = (1 - experienceFactor) * 0.5 * Math.cos((i - 288) * 0.06)
-  }
+/**
+ * Tool: kg.embeddings.similar
+ * Type: EntityId × Float → List[SimilarEntity]
+ */
+function embeddingsSimilar(embeddings, entityId, threshold = 0.7) {
+  // findSimilar returns JSON string with {entity, score, distance}
+  const resultJson = embeddings.findSimilar(entityId, 5, threshold)
+  const results = JSON.parse(resultJson)
+  // Map to standard format
+  return results.map(r => ({
+    id: r.entity,
+    similarity: r.score
+  }))
+}
 
-  // Normalize to unit vector
-  const magnitude = Math.sqrt(embedding.reduce((s, v) => s + v * v, 0))
-  if (magnitude > 0) {
-    for (let i = 0; i < 384; i++) embedding[i] /= magnitude
+/**
+ * Tool: kg.datalog.infer
+ * Type: DatalogProgram → InferredFacts
+ * Result format: {"predicate_name": [[term1, term2], ...], ...}
+ */
+function datalogInfer(program) {
+  const resultJson = evaluateDatalog(program)
+  const result = JSON.parse(resultJson)
+  // Convert to array format for easier processing
+  const facts = []
+  for (const [predicate, termArrays] of Object.entries(result)) {
+    for (const terms of termArrays) {
+      facts.push({ name: predicate, args: terms })
+    }
   }
-
-  return embedding
+  return facts
 }
 
-// ============================================================================
-// ISO PREMIUM CALCULATION (Industry Standard Formula)
-// Premium = Base Rate × Exposure × Territory Mod × Experience Mod × Loss Mod
-// ============================================================================
-
+/**
+ * Tool: kg.premium.calculate
+ * Type: RiskFactors → PremiumQuote
+ *
+ * ISO Premium Calculation Formula:
+ * Premium = Base Rate × Exposure × Territory Mod × Experience Mod × Loss Mod
+ */
 function calculatePremium(baseRate, exposure, territoryMod, lossRatio, yearsInBusiness) {
-  // Experience modification (NAIC standard breaks)
-  const experienceMod = yearsInBusiness >= 10 ? 0.90 :
-                        yearsInBusiness >= 5 ? 0.95 : 1.05
+  // Experience modification (tenure discount)
+  const experienceMod = yearsInBusiness >= 10 ? 0.90 : yearsInBusiness >= 5 ? 0.95 : 1.05
 
-  // Loss development factor (actuarial standard)
-  const lossMod = lossRatio < 0.30 ? 0.85 :  // Excellent
-                  lossRatio < 0.50 ? 1.00 :  // Average
-                  lossRatio < 0.70 ? 1.15 :  // Below average
-                  1.35                        // Poor - surcharge
+  // Loss modification (claims history)
+  const lossMod = lossRatio < 0.30 ? 0.85 :
+                  lossRatio < 0.50 ? 1.00 :
+                  lossRatio < 0.70 ? 1.15 : 1.35
 
   const premium = baseRate * exposure * territoryMod * experienceMod * lossMod
   return Math.round(premium * 100) / 100
 }
 
-// ============================================================================
-// HYPERMIND AGENT: Underwriting Pipeline
-//
-// Tool Composition (Category Theory):
-//   kg.sparql.query      : Query → BindingSet
-//   kg.graphframe.analyze : Graph → RiskCorrelation
-//   kg.embeddings.search : Entity → SimilarRiskProfiles
-//   kg.datalog.infer     : Rules → UnderwritingDecisions
-//
-// Each tool is a typed morphism. Composition is type-checked.
-// ============================================================================
-
-async function runUnderwritingAgent() {
-  console.log('═'.repeat(74))
+// ═══════════════════════════════════════════════════════════════════════════
+// MAIN UNDERWRITING AGENT
+// ═══════════════════════════════════════════════════════════════════════════
+
+async function main() {
+  console.log('═'.repeat(78))
   console.log('  HYPERMIND UNDERWRITING AGENT')
-  console.log('  rust-kgdb v' + getVersion() + ' | Neuro-Symbolic AI Framework')
-  console.log('═'.repeat(74))
+  console.log(`  rust-kgdb ${getVersion()} | Neuro-Symbolic AI Framework`)
+  console.log('═'.repeat(78))
   console.log()
   console.log('  Architecture: Type Theory + Category Theory + Proof Theory')
-  console.log('  Data Source:  ISO rating factors, NAIC guidelines, NAICS codes')
+  console.log('  Data Sources: ISO rating factors, NAIC guidelines, NAICS codes')
+  console.log(`  LLM Model:    ${MODEL}`)
   console.log()
 
-  // Track execution for audit witness
-  const executionWitness = {
-    agent: 'UnderwritingAgent',
-    startTime: new Date().toISOString(),
-    tools: [],
-    decisions: []
-  }
+  // ─────────────────────────────────────────────────────────────────────────
+  // PHASE 1: Initialize Knowledge Graph
+  // ─────────────────────────────────────────────────────────────────────────
 
-  // ═══════════════════════════════════════════════════════════════════════════
-  // TOOL 1: kg.sparql.query (Morphism: Query → BindingSet)
-  // ═══════════════════════════════════════════════════════════════════════════
-  console.log('┌─ TOOL: kg.sparql.query ─────────────────────────────────────────────┐')
-  console.log('│  Type: SPARQLQuery → BindingSet                                     │')
+  console.log('┌─ TOOL: kg.sparql.load ───────────────────────────────────────────────┐')
+  console.log('│  Type: TTL → Graph                                                   │')
   console.log('└─────────────────────────────────────────────────────────────────────┘')
 
-  const toolStart1 = Date.now()
   const db = new GraphDB('http://underwriting.org/kb')
-  db.loadTtl(UNDERWRITING_DATA_TTL, null)
+  db.loadTtl(UNDERWRITING_KB, 'http://underwriting.org/data')
+  const tripleCount = db.countTriples()
+  console.log(`  Knowledge Graph: ${tripleCount} triples loaded`)
+  console.log()
 
-  console.log(`  Knowledge Graph: ${db.countTriples()} triples loaded`)
-  console.log(`  Graph URI: ${db.getGraphUri()}`)
+  // ─────────────────────────────────────────────────────────────────────────
+  // PHASE 2: Spawn HyperMind Agent
+  // ─────────────────────────────────────────────────────────────────────────
+
+  console.log('┌─ AGENT: HyperMindAgent.spawn() ──────────────────────────────────────┐')
+  console.log('│  AgentSpec: {                                                        │')
+  console.log('│    name: "underwriter",                                              │')
+  console.log(`│    model: "${MODEL}",                                       │`)
+  console.log('│    tools: [kg.sparql.query, kg.datalog.apply, kg.embeddings.search],│')
+  console.log('│    tracing: true                                                     │')
+  console.log('│  }                                                                   │')
+  console.log('└─────────────────────────────────────────────────────────────────────┘')
 
-  executionWitness.tools.push({
-    name: 'kg.sparql.query',
-    input: 'underwriting_knowledge_base',
-    output: `${db.countTriples()} triples`,
-    durationMs: Date.now() - toolStart1
+  const agent = await HyperMindAgent.spawn({
+    name: 'underwriter',
+    model: MODEL,
+    tools: ['kg.sparql.query', 'kg.datalog.apply', 'kg.embeddings.search'],
+    endpoint: 'http://localhost:30080',
+    tracing: true
   })
+
+  console.log(`  Agent "${agent.getName()}" spawned with model: ${agent.getModel()}`)
   console.log()
 
-  // ═══════════════════════════════════════════════════════════════════════════
-  // TOOL 2: kg.graphframe.analyze (Morphism: Graph → RiskCorrelation)
-  // ═══════════════════════════════════════════════════════════════════════════
-  console.log('┌─ TOOL: kg.graphframe.analyze ───────────────────────────────────────┐')
-  console.log('│  Type: Graph → {pagerank, components, correlations}                 │')
+  // ─────────────────────────────────────────────────────────────────────────
+  // PHASE 3: Execute Underwriting Tools
+  // ─────────────────────────────────────────────────────────────────────────
+
+  // Tool 1: SPARQL Query - Get All Accounts
+  console.log('┌─ TOOL: kg.sparql.query ─────────────────────────────────────────────┐')
+  console.log('│  Type: SPARQLQuery → BindingSet                                     │')
   console.log('└─────────────────────────────────────────────────────────────────────┘')
 
-  const toolStart2 = Date.now()
-
-  // Build risk correlation graph
-  const vertices = JSON.stringify([
-    { id: 'BUS001', type: 'account' },
-    { id: 'BUS002', type: 'account' },
-    { id: 'BUS003', type: 'account' },
-    { id: 'BUS004', type: 'account' },
-    { id: 'hurricane', type: 'catastrophe' },
-    { id: 'earthquake', type: 'catastrophe' },
-    { id: 'tornado', type: 'catastrophe' },
-    { id: 'flood', type: 'catastrophe' },
-    { id: 'manufacturing', type: 'industry' },
-    { id: 'technology', type: 'industry' },
-    { id: 'transportation', type: 'industry' },
-    { id: 'hospitality', type: 'industry' }
-  ])
-
-  const edges = JSON.stringify([
-    // Business to CAT exposure
-    { src: 'BUS001', dst: 'hurricane', type: 'exposure' },
-    { src: 'BUS002', dst: 'earthquake', type: 'exposure' },
-    { src: 'BUS003', dst: 'tornado', type: 'exposure' },
-    { src: 'BUS004', dst: 'flood', type: 'exposure' },
-    // Business to industry class
-    { src: 'BUS001', dst: 'manufacturing', type: 'class' },
-    { src: 'BUS002', dst: 'technology', type: 'class' },
-    { src: 'BUS003', dst: 'transportation', type: 'class' },
-    { src: 'BUS004', dst: 'hospitality', type: 'class' },
-    // Correlated risks (for portfolio analysis)
-    { src: 'hurricane', dst: 'flood', type: 'correlation' },
-    { src: 'manufacturing', dst: 'transportation', type: 'supplychain' }
-  ])
-
-  const graph = new GraphFrame(vertices, edges)
-
-  // PageRank for risk concentration analysis
-  const pageRankResult = JSON.parse(graph.pageRank(0.15, 20))
-
-  // Connected components for portfolio clustering
-  const components = JSON.parse(graph.connectedComponents())
-
-  executionWitness.tools.push({
-    name: 'kg.graphframe.analyze',
-    input: `${graph.vertexCount()} vertices, ${graph.edgeCount()} edges`,
-    output: 'risk_correlation_complete',
-    durationMs: Date.now() - toolStart2
+  const accountsQuery = `
+    PREFIX uw: <http://underwriting.org/>
+    PREFIX xsd: <http://www.w3.org/2001/XMLSchema#>
+    SELECT ?account ?name ?industry ?revenue ?lossRatio ?years WHERE {
+      ?account a uw:BusinessAccount ;
+               uw:name ?name ;
+               uw:industry ?industry ;
+               uw:annualRevenue ?revenue ;
+               uw:lossRatio ?lossRatio ;
+               uw:yearsInBusiness ?years .
+    }
+  `
+  const accounts = await sparqlQuery(db, accountsQuery)
+  console.log(`  Accounts retrieved: ${accounts.length}`)
+  accounts.forEach(a => {
+    const name = a['?name'] || a.name
+    const lr = a['?lossRatio'] || a.lossRatio
+    console.log(`    - ${name}: loss ratio ${lr}`)
   })
-
-  console.log(`  Risk Network: ${graph.vertexCount()} nodes, ${graph.edgeCount()} edges`)
-  console.log('  Risk Concentration (PageRank):')
-  if (pageRankResult.ranks) {
-    Object.entries(pageRankResult.ranks)
-      .filter(([k]) => k.startsWith('BUS'))
-      .sort((a, b) => b[1] - a[1])
-      .forEach(([node, score]) => {
-        console.log(`    - ${node}: ${score.toFixed(4)}`)
-      })
-  }
   console.log()
 
-  // ═══════════════════════════════════════════════════════════════════════════
-  // TOOL 3: kg.embeddings.search (Morphism: Entity → SimilarRiskProfiles)
-  // ═══════════════════════════════════════════════════════════════════════════
-  console.log('┌─ TOOL: kg.embeddings.search ────────────────────────────────────────┐')
-  console.log('│  Type: EntityId → List[{entity, similarity}]                        │')
+  // Tool 2: GraphFrame - Risk Network Analysis
+  console.log('┌─ TOOL: kg.graphframe.analyze ────────────────────────────────────────┐')
+  console.log('│  Type: Graph → {pagerank, components}                               │')
   console.log('└─────────────────────────────────────────────────────────────────────┘')
 
-  const toolStart3 = Date.now()
-
-  const embeddings = new EmbeddingService()
+  const riskVertices = [
+    { id: 'BUS001', type: 'account', industry: 'manufacturing' },
+    { id: 'BUS002', type: 'account', industry: 'technology' },
+    { id: 'BUS003', type: 'account', industry: 'transportation' },
+    { id: 'BUS004', type: 'account', industry: 'hospitality' },
+    { id: 'FL', type: 'territory' },
+    { id: 'CA', type: 'territory' },
+    { id: 'TX', type: 'territory' },
+    { id: 'NY', type: 'territory' }
+  ]
 
-  // Risk profiles based on real underwriting factors
-  const riskProfiles = [
-    { id: 'BUS001', class: 'manufacturing', lossRatio: 0.45, territory: 1.35, years: 15 },
-    { id: 'BUS002', class: 'technology', lossRatio: 0.00, territory: 1.15, years: 3 },
-    { id: 'BUS003', class: 'transportation', lossRatio: 0.72, territory: 1.05, years: 8 },
-    { id: 'BUS004', class: 'hospitality', lossRatio: 0.28, territory: 1.25, years: 12 }
+  const riskEdges = [
+    { src: 'BUS001', dst: 'FL', relationship: 'located_in' },
+    { src: 'BUS002', dst: 'CA', relationship: 'located_in' },
+    { src: 'BUS003', dst: 'TX', relationship: 'located_in' },
+    { src: 'BUS004', dst: 'NY', relationship: 'located_in' },
+    { src: 'FL', dst: 'TX', relationship: 'hurricane_corridor' },
+    { src: 'BUS001', dst: 'BUS003', relationship: 'same_class' }
   ]
 
-  riskProfiles.forEach(p => {
-    const vec = generateRiskEmbedding(p.class, p.lossRatio, p.territory, p.years)
-    embeddings.storeVector(p.id, vec)
+  const analysis = graphframeAnalyze(riskVertices, riskEdges)
+  console.log(`  Risk Network: ${riskVertices.length} nodes, ${riskEdges.length} edges`)
+  console.log('  Risk Concentration (PageRank):')
+  Object.entries(analysis.pageRank).slice(0, 4).forEach(([node, rank]) => {
+    console.log(`    - ${node}: ${rank.toFixed(4)}`)
   })
+  console.log()
 
-  embeddings.rebuildIndex()
+  // Tool 3: Embeddings - Similar Risk Profiles
+  console.log('┌─ TOOL: kg.embeddings.similar ───────────────────────────────────────┐')
+  console.log('│  Type: EntityId × Float → List[SimilarEntity]                       │')
+  console.log('└─────────────────────────────────────────────────────────────────────┘')
 
-  // Find accounts with similar risk profiles to BUS003 (highest risk)
-  const similarProfiles = JSON.parse(embeddings.findSimilar('BUS003', 5, 0.3))
+  const embeddings = new EmbeddingService()
 
-  executionWitness.tools.push({
-    name: 'kg.embeddings.search',
-    input: 'BUS003 (high-risk transportation)',
-    output: `${similarProfiles.length} similar profiles`,
-    durationMs: Date.now() - toolStart3
-  })
+  // Add account embeddings based on risk characteristics
+  embeddings.storeVector('BUS001', generateRiskEmbedding('manufacturing', 0.45, 15, 1.35))
+  embeddings.storeVector('BUS002', generateRiskEmbedding('technology', 0.15, 3, 1.25))
+  embeddings.storeVector('BUS003', generateRiskEmbedding('transportation', 0.72, 8, 1.20))
+  embeddings.storeVector('BUS004', generateRiskEmbedding('hospitality', 0.28, 12, 1.30))
 
-  console.log(`  Risk Embeddings: ${riskProfiles.length} accounts vectorized (384 dims)`)
+  const similarAccounts = embeddingsSimilar(embeddings, 'BUS003', 0.5)
   console.log('  Accounts similar to BUS003 (highest risk):')
-  similarProfiles
-    .filter(s => s.entity !== 'BUS003')
-    .forEach(s => {
-      const profile = riskProfiles.find(p => p.id === s.entity)
-      if (profile) {
-        console.log(`    - ${s.entity}: ${profile.class}, loss ratio ${profile.lossRatio}`)
-      }
-    })
+  similarAccounts.forEach(s => {
+    console.log(`    - ${s.id}: similarity ${s.similarity.toFixed(3)}`)
+  })
   console.log()
 
-  // ═══════════════════════════════════════════════════════════════════════════
-  // TOOL 4: kg.datalog.infer (Morphism: Rules → UnderwritingDecisions)
-  // ═══════════════════════════════════════════════════════════════════════════
+  // Tool 4: Datalog - Underwriting Decision Rules
   console.log('┌─ TOOL: kg.datalog.infer ────────────────────────────────────────────┐')
   console.log('│  Type: DatalogProgram → UnderwritingDecisions                       │')
   console.log('└─────────────────────────────────────────────────────────────────────┘')
 
-  const toolStart4 = Date.now()
-
   const datalog = new DatalogProgram()
 
-  // Facts: Business profiles
-  datalog.addFact(JSON.stringify({ predicate: 'business', terms: ['BUS001', 'manufacturing', '0.45'] }))
-  datalog.addFact(JSON.stringify({ predicate: 'business', terms: ['BUS002', 'technology', '0.00'] }))
-  datalog.addFact(JSON.stringify({ predicate: 'business', terms: ['BUS003', 'transportation', '0.72'] }))
-  datalog.addFact(JSON.stringify({ predicate: 'business', terms: ['BUS004', 'hospitality', '0.28'] }))
-
-  // High-risk class definitions
-  datalog.addFact(JSON.stringify({ predicate: 'highRiskClass', terms: ['transportation'] }))
-  datalog.addFact(JSON.stringify({ predicate: 'highRiskClass', terms: ['construction'] }))
-
-  // Rule: Refer to underwriter if high-risk class
-  // referToUW(Bus) :- business(Bus, Class, _), highRiskClass(Class)
+  // Facts based on account characteristics - using correct JSON API
+  datalog.addFact(JSON.stringify({ predicate: 'account', terms: ['BUS001'] }))
+  datalog.addFact(JSON.stringify({ predicate: 'account', terms: ['BUS002'] }))
+  datalog.addFact(JSON.stringify({ predicate: 'account', terms: ['BUS003'] }))
+  datalog.addFact(JSON.stringify({ predicate: 'account', terms: ['BUS004'] }))
+  datalog.addFact(JSON.stringify({ predicate: 'low_loss_ratio', terms: ['BUS002'] }))  // < 0.30
+  datalog.addFact(JSON.stringify({ predicate: 'low_loss_ratio', terms: ['BUS004'] }))  // < 0.30
+  datalog.addFact(JSON.stringify({ predicate: 'high_loss_ratio', terms: ['BUS003'] })) // > 0.70
+  datalog.addFact(JSON.stringify({ predicate: 'established', terms: ['BUS001'] }))     // > 10 years
+  datalog.addFact(JSON.stringify({ predicate: 'established', terms: ['BUS004'] }))
+  datalog.addFact(JSON.stringify({ predicate: 'high_risk_class', terms: ['BUS003'] })) // transportation
+
+  // NAIC-informed underwriting rules - using correct JSON API
   datalog.addRule(JSON.stringify({
-    head: { predicate: 'referToUW', terms: ['?Bus'] },
+    head: { predicate: 'auto_approve', terms: ['?X'] },
     body: [
-      { predicate: 'business', terms: ['?Bus', '?Class', '?LR'] },
-      { predicate: 'highRiskClass', terms: ['?Class'] }
+      { predicate: 'account', terms: ['?X'] },
+      { predicate: 'low_loss_ratio', terms: ['?X'] }
     ]
   }))
 
-  // Rule: Auto-approve if technology class (low risk)
-  // autoApprove(Bus) :- business(Bus, technology, _)
   datalog.addRule(JSON.stringify({
-    head: { predicate: 'autoApprove', terms: ['?Bus'] },
+    head: { predicate: 'refer_to_underwriter', terms: ['?X'] },
     body: [
-      { predicate: 'business', terms: ['?Bus', 'technology', '?LR'] }
+      { predicate: 'account', terms: ['?X'] },
+      { predicate: 'high_loss_ratio', terms: ['?X'] }
     ]
   }))
 
-  const inferredResult = evaluateDatalog(datalog)
-  const inferred = JSON.parse(inferredResult)
+  const decisions = datalogInfer(datalog)
+  const autoApprove = decisions.filter(d => d.name === 'auto_approve')
+  const referToUW = decisions.filter(d => d.name === 'refer_to_underwriter')
 
-  executionWitness.tools.push({
-    name: 'kg.datalog.infer',
-    input: `${datalog.factCount()} facts, ${datalog.ruleCount()} rules`,
-    output: `autoApprove=${inferred.autoApprove?.length || 0}, referToUW=${inferred.referToUW?.length || 0}`,
-    durationMs: Date.now() - toolStart4
-  })
-
-  console.log(`  Facts: ${datalog.factCount()}, Rules: ${datalog.ruleCount()}`)
-  console.log('  Decisions:')
-  if (inferred.autoApprove && inferred.autoApprove.length > 0) {
-    inferred.autoApprove.forEach(a => {
-      console.log(`    ✅ AUTO-APPROVE: ${a[0]}`)
-      executionWitness.decisions.push({ account: a[0], decision: 'AUTO-APPROVE' })
-    })
-  }
-  if (inferred.referToUW && inferred.referToUW.length > 0) {
-    inferred.referToUW.forEach(r => {
-      console.log(`    ⚠️  REFER TO UW: ${r[0]}`)
-      executionWitness.decisions.push({ account: r[0], decision: 'REFER-TO-UW' })
-    })
-  }
+  console.log(`  Inferred decisions: ${decisions.length}`)
+  console.log(`  Auto-approve: ${autoApprove.length}`)
+  autoApprove.forEach(d => console.log(`    - ${d.args[0]}`))
+  console.log(`  Refer to underwriter: ${referToUW.length}`)
+  referToUW.forEach(d => console.log(`    - ${d.args[0]}`))
   console.log()
 
-  // ═══════════════════════════════════════════════════════════════════════════
-  // PREMIUM CALCULATION (ISO Standard Methodology)
-  // ═══════════════════════════════════════════════════════════════════════════
-  console.log('┌─ TOOL: kg.premium.calculate ────────────────────────────────────────┐')
-  console.log('│  Formula: Base × Exposure × Territory × Experience × Loss          │')
+  // Tool 5: Premium Calculation
+  console.log('┌─ TOOL: kg.premium.calculate ─────────────────────────────────────────┐')
+  console.log('│  Formula: Base × Exposure × Territory × Experience × Loss           │')
   console.log('└─────────────────────────────────────────────────────────────────────┘')
 
-  const premiumData = [
-    { id: 'BUS001', name: 'Acme Manufacturing', baseRate: 2.45, exposure: 45000000, territory: 1.35, loss: 0.45, years: 15 },
-    { id: 'BUS002', name: 'TechStart LLC', baseRate: 0.85, exposure: 8500000, territory: 1.15, loss: 0.00, years: 3 },
-    { id: 'BUS003', name: 'SafeHaul Logistics', baseRate: 3.80, exposure: 22000000, territory: 1.05, loss: 0.72, years: 8 },
-    { id: 'BUS004', name: 'Downtown Restaurant', baseRate: 1.95, exposure: 12000000, territory: 1.25, loss: 0.28, years: 12 }
+  const premiums = [
+    {
+      id: 'BUS001',
+      name: 'Acme Manufacturing',
+      premium: calculatePremium(12.50, 8500000 / 100, 1.35, 0.45, 15),
+      decision: 'STANDARD'
+    },
+    {
+      id: 'BUS002',
+      name: 'TechStart LLC',
+      premium: calculatePremium(4.25, 1200000 / 100, 1.25, 0.15, 3),
+      decision: 'APPROVED'
+    },
+    {
+      id: 'BUS003',
+      name: 'SafeHaul Logistics',
+      premium: calculatePremium(18.75, 4200000 / 100, 1.45, 0.72, 8),
+      decision: 'REFER'
+    },
+    {
+      id: 'BUS004',
+      name: 'Downtown Restaurant',
+      premium: calculatePremium(8.90, 3100000 / 100, 1.30, 0.28, 12),
+      decision: 'STANDARD'
+    }
   ]
 
   console.log('  Calculated Premiums:')
-  premiumData.forEach(p => {
-    const premium = calculatePremium(p.baseRate, p.exposure / 100, p.territory, p.loss, p.years)
-    const decision = inferred.autoApprove?.some(a => a[0] === p.id) ? 'APPROVED' :
-                     inferred.referToUW?.some(r => r[0] === p.id) ? 'REFER' : 'STANDARD'
-    console.log(`    - ${p.id} (${p.name}): $${premium.toLocaleString()} [${decision}]`)
+  premiums.forEach(p => {
+    console.log(`    - ${p.id} (${p.name}): $${p.premium.toLocaleString()} [${p.decision}]`)
   })
   console.log()
 
-  // ═══════════════════════════════════════════════════════════════════════════
-  // EXECUTION WITNESS (Proof Theory - Curry-Howard Correspondence)
-  // ═══════════════════════════════════════════════════════════════════════════
-  executionWitness.endTime = new Date().toISOString()
-  executionWitness.summary = {
-    accountsProcessed: riskProfiles.length,
-    autoApproved: inferred.autoApprove?.length || 0,
-    referredToUW: inferred.referToUW?.length || 0,
-    standardProcessing: riskProfiles.length - (inferred.autoApprove?.length || 0) - (inferred.referToUW?.length || 0)
-  }
+  // ─────────────────────────────────────────────────────────────────────────
+  // PHASE 4: Agent LLM Call (if real API key available)
+  // ─────────────────────────────────────────────────────────────────────────
 
-  // Generate deterministic hash of execution
-  const witnessString = JSON.stringify(executionWitness)
-  let hash = 0
-  for (let i = 0; i < witnessString.length; i++) {
-    const char = witnessString.charCodeAt(i)
-    hash = ((hash << 5) - hash) + char
-    hash = hash & hash
+  if (MODEL !== 'mock') {
+    console.log('┌─ AGENT: agent.call() ─────────────────────────────────────────────┐')
+    console.log('│  Natural Language → SPARQL → Execution                            │')
+    console.log('└─────────────────────────────────────────────────────────────────────┘')
+
+    try {
+      const result = await agent.call('Find all business accounts with loss ratio above 0.5')
+      console.log(`  LLM Generated SPARQL: ${result.success ? 'SUCCESS' : 'FAILED'}`)
+      if (result.sparql) {
+        console.log(`  Query: ${result.sparql.slice(0, 60)}...`)
+      }
+      if (result.error) {
+        console.log(`  Error: ${result.error}`)
+      }
+    } catch (e) {
+      console.log(`  Agent call skipped: ${e.message}`)
+    }
+    console.log()
   }
-  executionWitness.hash = 'sha256:' + Math.abs(hash).toString(16).padStart(16, '0')
 
-  // ═══════════════════════════════════════════════════════════════════════════
-  // FINAL REPORT
-  // ═══════════════════════════════════════════════════════════════════════════
-  console.log('═'.repeat(74))
+  // ─────────────────────────────────────────────────────────────────────────
+  // PHASE 5: Generate Execution Witness (Proof Theory)
+  // ─────────────────────────────────────────────────────────────────────────
+
+  console.log('═'.repeat(78))
   console.log('  UNDERWRITING DECISION REPORT')
-  console.log('═'.repeat(74))
+  console.log('═'.repeat(78))
   console.log()
+
   console.log('  SUMMARY:')
-  console.log(`    Accounts processed:     ${executionWitness.summary.accountsProcessed}`)
-  console.log(`    Auto-approved:          ${executionWitness.summary.autoApproved}`)
-  console.log(`    Referred to UW:         ${executionWitness.summary.referredToUW}`)
-  console.log(`    Standard processing:    ${executionWitness.summary.standardProcessing}`)
+  console.log(`    Accounts processed:     ${accounts.length}`)
+  console.log(`    Auto-approved:          ${autoApprove.length}`)
+  console.log(`    Referred to UW:         ${referToUW.length}`)
+  console.log(`    Standard processing:    ${accounts.length - autoApprove.length - referToUW.length}`)
   console.log()
+
   console.log('  RISK INDICATORS:')
   console.log('    - BUS003 (SafeHaul): HIGH RISK - 72% loss ratio, transportation class')
   console.log('    - BUS001 (Acme): MODERATE - hurricane exposure, 45% loss ratio')
   console.log('    - BUS002 (TechStart): LOW - clean history, technology class')
   console.log('    - BUS004 (Downtown): MODERATE - flood exposure, established')
   console.log()
+
+  // Execution Witness
+  const witness = {
+    agent: agent.getName(),
+    model: agent.getModel(),
+    timestamp: new Date().toISOString(),
+    tools: [
+      { name: 'kg.sparql.query', input: 'accounts', output: accounts.length },
+      { name: 'kg.graphframe.analyze', input: `${riskVertices.length} nodes`, output: 'pagerank_complete' },
+      { name: 'kg.embeddings.similar', input: 'BUS003', output: similarAccounts.length },
+      { name: 'kg.datalog.infer', input: '10 facts, 2 rules', output: decisions.length },
+      { name: 'kg.premium.calculate', input: '4 accounts', output: premiums.length }
+    ],
+    decisions: {
+      autoApprove: autoApprove.map(d => d.args[0]),
+      referToUW: referToUW.map(d => d.args[0])
+    },
+    premiums: premiums.map(p => ({ id: p.id, premium: p.premium, decision: p.decision })),
+    trace: agent.getTrace()
+  }
+
+  // Generate deterministic hash
+  const hashInput = JSON.stringify(witness.decisions) + JSON.stringify(witness.premiums)
+  let hash = 0
+  for (let i = 0; i < hashInput.length; i++) {
+    hash = ((hash << 5) - hash) + hashInput.charCodeAt(i)
+    hash |= 0
+  }
+  witness.hash = 'sha256:' + Math.abs(hash).toString(16).padStart(16, '0')
+
   console.log('  EXECUTION WITNESS (for audit):')
-  console.log(`    Hash: ${executionWitness.hash}`)
-  console.log(`    Tools executed: ${executionWitness.tools.length}`)
-  console.log(`    Duration: ${executionWitness.tools.reduce((s, t) => s + t.durationMs, 0)}ms`)
+  console.log(`    Hash: ${witness.hash}`)
+  console.log(`    Agent: ${witness.agent}`)
+  console.log(`    Model: ${witness.model}`)
+  console.log(`    Tools executed: ${witness.tools.length}`)
+  console.log()
+  console.log('═'.repeat(78))
+  console.log()
+  console.log('HyperMind Agent completed successfully.')
   console.log()
-  console.log('═'.repeat(74))
+  console.log('Full witness:', JSON.stringify(witness, null, 2))
+}
 
-  return executionWitness
+/**
+ * Generate a risk embedding vector based on account characteristics
+ */
+function generateRiskEmbedding(industry, lossRatio, yearsInBusiness, territoryMod) {
+  const embedding = new Float32Array(384)
+
+  // Industry encoding
+  const industries = {
+    manufacturing: 0.5,
+    technology: 0.2,
+    transportation: 0.8,
+    hospitality: 0.4
+  }
+  embedding[0] = industries[industry] || 0.5
+
+  // Loss ratio (already 0-1)
+  embedding[1] = lossRatio
+
+  // Years normalized (0-30 range)
+  embedding[2] = Math.min(yearsInBusiness / 30, 1.0)
+
+  // Territory risk
+  embedding[3] = (territoryMod - 1.0) * 2  // Normalize 1.0-1.5 to 0-1
+
+  // Fill rest with deterministic values
+  for (let i = 4; i < 384; i++) {
+    embedding[i] = ((embedding[0] * embedding[1] * (i + 1)) % 1)
+  }
+
+  return Array.from(embedding)
 }
 
-// Execute agent
-runUnderwritingAgent()
-  .then(witness => {
-    console.log('\nHyperMind Agent completed successfully.')
-    console.log('Full witness:', JSON.stringify(witness, null, 2))
-    process.exit(0)
-  })
-  .catch(err => {
-    console.error('Agent failed:', err.message)
-    process.exit(1)
-  })
+// Run
+main().catch(console.error)