@ruvector/ruvllm 2.5.6 → 2.5.7

package/scripts/hybrid-model-compare.js

@@ -1,373 +1,373 @@
-#!/usr/bin/env node
-/**
- * Hybrid Model Comparison
- *
- * Combines embedding similarity with keyword boosting.
- * This addresses the "reviewer overfit" problem by:
- * 1. Computing embedding similarity
- * 2. Boosting agents that have keyword matches in the task
- * 3. Using weighted combination for final score
- */
-
-const { execSync } = require('child_process');
-const { existsSync } = require('fs');
-const { join } = require('path');
-const { homedir } = require('os');
-
-const MODELS_DIR = join(homedir(), '.ruvllm', 'models');
-const RUVLTRA_MODEL = join(MODELS_DIR, 'ruvltra-claude-code-0.5b-q4_k_m.gguf');
-const QWEN_MODEL = join(MODELS_DIR, 'qwen2.5-0.5b-instruct-q4_k_m.gguf');
-
-// V1 descriptions for embedding
-const DESCRIPTIONS_V1 = {
-  coder: 'implement create write build add code function class component feature',
-  researcher: 'research find investigate analyze explore search discover examine',
-  reviewer: 'review check evaluate assess inspect examine code quality',
-  tester: 'test unit integration e2e coverage mock assertion spec',
-  architect: 'design architecture schema system structure plan database',
-  'security-architect': 'security vulnerability xss injection audit cve authentication',
-  debugger: 'debug fix bug error issue broken crash exception trace',
-  documenter: 'document readme jsdoc comment explain describe documentation',
-  refactorer: 'refactor extract rename consolidate clean restructure simplify',
-  optimizer: 'optimize performance slow fast cache speed memory latency',
-  devops: 'deploy ci cd kubernetes docker pipeline container infrastructure',
-  'api-docs': 'openapi swagger api documentation graphql schema endpoint',
-  planner: 'plan estimate prioritize sprint roadmap schedule milestone',
-};
-
-// UNIQUE trigger keywords - words that strongly indicate a specific agent
-// Priority-ordered: first match wins for disambiguation
-// NOTE: "investigate" takes priority over "slow" for researcher vs optimizer
-const TRIGGER_KEYWORDS = {
-  // Higher priority agents (check these first)
-  researcher: ['research', 'investigate', 'explore', 'discover', 'best practices', 'patterns', 'analyze', 'look into', 'find out'],
-  coder: ['implement', 'build', 'create', 'component', 'function', 'typescript', 'react', 'feature', 'write code'],
-  tester: ['test', 'tests', 'testing', 'unit test', 'integration test', 'e2e', 'coverage', 'spec'],
-  reviewer: ['review', 'pull request', 'pr', 'code quality', 'code review', 'check code'],
-  debugger: ['debug', 'fix', 'bug', 'error', 'exception', 'crash', 'trace', 'null pointer', 'memory leak'],
-  'security-architect': ['security', 'vulnerability', 'xss', 'injection', 'csrf', 'cve', 'audit', 'exploit'],
-  refactorer: ['refactor', 'async/await', 'modernize', 'restructure', 'extract', 'legacy'],
-  // Optimizer: removed "slow" (too generic), added query-specific terms
-  optimizer: ['optimize', 'performance', 'cache', 'caching', 'speed up', 'latency', 'faster', 'queries', 'reduce time'],
-  architect: ['design', 'architecture', 'schema', 'structure', 'diagram', 'system design', 'plan architecture'],
-  documenter: ['jsdoc', 'comment', 'comments', 'readme', 'documentation', 'document', 'explain'],
-  devops: ['deploy', 'ci/cd', 'kubernetes', 'docker', 'pipeline', 'infrastructure', 'container'],
-  'api-docs': ['openapi', 'swagger', 'api doc', 'rest api', 'graphql', 'endpoint'],
-  planner: ['sprint', 'plan', 'roadmap', 'milestone', 'estimate', 'schedule', 'prioritize'],
-};
-
-// Priority order for disambiguation (when multiple agents match)
-const AGENT_PRIORITY = [
-  'researcher',     // "investigate" wins over "slow"
-  'debugger',       // "fix" wins over generic terms
-  'tester',         // "test" is specific
-  'security-architect',
-  'coder',
-  'reviewer',
-  'refactorer',
-  'optimizer',
-  'architect',
-  'documenter',
-  'devops',
-  'api-docs',
-  'planner',
-];
-
-const ROUTING_TESTS = [
-  { task: 'Implement a binary search function in TypeScript', expected: 'coder' },
-  { task: 'Write unit tests for the authentication module', expected: 'tester' },
-  { task: 'Review the pull request for security vulnerabilities', expected: 'reviewer' },
-  { task: 'Research best practices for React state management', expected: 'researcher' },
-  { task: 'Design the database schema for user profiles', expected: 'architect' },
-  { task: 'Fix the null pointer exception in the login handler', expected: 'debugger' },
-  { task: 'Audit the API endpoints for XSS vulnerabilities', expected: 'security-architect' },
-  { task: 'Write JSDoc comments for the utility functions', expected: 'documenter' },
-  { task: 'Refactor the payment module to use async/await', expected: 'refactorer' },
-  { task: 'Optimize the database queries for the dashboard', expected: 'optimizer' },
-  { task: 'Set up the CI/CD pipeline for the microservices', expected: 'devops' },
-  { task: 'Generate OpenAPI documentation for the REST API', expected: 'api-docs' },
-  { task: 'Create a sprint plan for the next two weeks', expected: 'planner' },
-  { task: 'Build a React component for user registration', expected: 'coder' },
-  { task: 'Debug memory leak in the WebSocket handler', expected: 'debugger' },
-  { task: 'Investigate slow API response times', expected: 'researcher' },
-  { task: 'Check code for potential race conditions', expected: 'reviewer' },
-  { task: 'Add integration tests for the payment gateway', expected: 'tester' },
-  { task: 'Plan the architecture for real-time notifications', expected: 'architect' },
-  { task: 'Cache the frequently accessed user data', expected: 'optimizer' },
-];
-
-function getEmbedding(modelPath, text) {
-  try {
-    const sanitized = text.replace(/"/g, '\\"').replace(/\n/g, ' ');
-    const result = execSync(
-      `llama-embedding -m "${modelPath}" -p "${sanitized}" --embd-output-format json 2>/dev/null`,
-      { encoding: 'utf-8', maxBuffer: 10 * 1024 * 1024 }
-    );
-    const json = JSON.parse(result);
-    return json.data[json.data.length - 1].embedding;
-  } catch {
-    return null;
-  }
-}
-
-function cosineSimilarity(a, b) {
-  if (!a || !b || a.length !== b.length) return 0;
-  let dot = 0, normA = 0, normB = 0;
-  for (let i = 0; i < a.length; i++) {
-    dot += a[i] * b[i];
-    normA += a[i] * a[i];
-    normB += b[i] * b[i];
-  }
-  return dot / (Math.sqrt(normA) * Math.sqrt(normB) || 1);
-}
-
-/**
- * Count keyword matches for each agent
- */
-function getKeywordScores(task) {
-  const taskLower = task.toLowerCase();
-  const scores = {};
-
-  for (const [agent, keywords] of Object.entries(TRIGGER_KEYWORDS)) {
-    let matches = 0;
-    for (const kw of keywords) {
-      if (taskLower.includes(kw.toLowerCase())) {
-        matches++;
-      }
-    }
-    scores[agent] = matches;
-  }
-
-  return scores;
-}
-
-/**
- * Pure embedding routing (baseline)
- */
-function routeEmbeddingOnly(taskEmbedding, agentEmbeddings) {
-  let bestAgent = 'coder';
-  let bestSim = -1;
-
-  for (const [agent, emb] of Object.entries(agentEmbeddings)) {
-    const sim = cosineSimilarity(taskEmbedding, emb);
-    if (sim > bestSim) {
-      bestSim = sim;
-      bestAgent = agent;
-    }
-  }
-
-  return { agent: bestAgent, confidence: bestSim };
-}
-
-/**
- * Pure keyword routing
- */
-function routeKeywordOnly(task) {
-  const scores = getKeywordScores(task);
-  let bestAgent = 'coder';
-  let bestScore = 0;
-
-  for (const [agent, score] of Object.entries(scores)) {
-    if (score > bestScore) {
-      bestScore = score;
-      bestAgent = agent;
-    }
-  }
-
-  return { agent: bestAgent, confidence: bestScore };
-}
-
-/**
- * Hybrid routing - combine embedding similarity with keyword boost
- */
-function routeHybrid(task, taskEmbedding, agentEmbeddings, embeddingWeight = 0.6, keywordWeight = 0.4) {
-  const keywordScores = getKeywordScores(task);
-
-  // Normalize keyword scores to 0-1 range
-  const maxKeyword = Math.max(...Object.values(keywordScores), 1);
-  const normalizedKeywords = {};
-  for (const agent of Object.keys(keywordScores)) {
-    normalizedKeywords[agent] = keywordScores[agent] / maxKeyword;
-  }
-
-  let bestAgent = 'coder';
-  let bestScore = -1;
-  const allScores = {};
-
-  for (const [agent, emb] of Object.entries(agentEmbeddings)) {
-    const embSim = cosineSimilarity(taskEmbedding, emb);
-    const kwScore = normalizedKeywords[agent] || 0;
-    const combined = embeddingWeight * embSim + keywordWeight * kwScore;
-    allScores[agent] = { embedding: embSim, keyword: kwScore, combined };
-
-    if (combined > bestScore) {
-      bestScore = combined;
-      bestAgent = agent;
-    }
-  }
-
-  return { agent: bestAgent, confidence: bestScore, scores: allScores };
-}
-
-/**
- * Keyword-first routing - use keywords as primary, embedding as tiebreaker
- */
-function routeKeywordFirst(task, taskEmbedding, agentEmbeddings) {
-  const keywordScores = getKeywordScores(task);
-
-  // Find agents with max keyword matches
-  const maxKw = Math.max(...Object.values(keywordScores));
-
-  if (maxKw > 0) {
-    // At least one keyword match - use keywords, embedding as tiebreaker
-    const candidates = Object.entries(keywordScores)
-      .filter(([_, score]) => score === maxKw)
-      .map(([agent, _]) => agent);
-
-    if (candidates.length === 1) {
-      return { agent: candidates[0], confidence: maxKw };
-    }
-
-    // Multiple candidates with same keyword count - use embedding
-    let bestAgent = candidates[0];
-    let bestSim = -1;
-    for (const agent of candidates) {
-      const sim = cosineSimilarity(taskEmbedding, agentEmbeddings[agent]);
-      if (sim > bestSim) {
-        bestSim = sim;
-        bestAgent = agent;
-      }
-    }
-    return { agent: bestAgent, confidence: maxKw + bestSim / 10 };
-  }
-
-  // No keyword matches - fall back to pure embedding
-  return routeEmbeddingOnly(taskEmbedding, agentEmbeddings);
-}
-
-function runBenchmark(modelPath, routerFn, name) {
-  const agentEmbeddings = {};
-  for (const [agent, desc] of Object.entries(DESCRIPTIONS_V1)) {
-    agentEmbeddings[agent] = getEmbedding(modelPath, desc);
-  }
-
-  let correct = 0;
-  const results = [];
-
-  for (const test of ROUTING_TESTS) {
-    const taskEmb = getEmbedding(modelPath, test.task);
-    const { agent } = routerFn(test.task, taskEmb, agentEmbeddings);
-    const isCorrect = agent === test.expected;
-    if (isCorrect) correct++;
-    results.push({ task: test.task, expected: test.expected, got: agent, correct: isCorrect });
-  }
-
-  return { accuracy: correct / ROUTING_TESTS.length, correct, total: ROUTING_TESTS.length, results, name };
-}
-
-async function main() {
-  console.log('\n╔═══════════════════════════════════════════════════════════════════════════════════╗');
-  console.log('║                 HYBRID ROUTING: Embeddings + Keywords                             ║');
-  console.log('╚═══════════════════════════════════════════════════════════════════════════════════╝\n');
-
-  if (!existsSync(RUVLTRA_MODEL)) {
-    console.error('RuvLTRA model not found.');
-    process.exit(1);
-  }
-
-  console.log('Strategies:');
-  console.log('  1. Embedding Only (baseline)');
-  console.log('  2. Keyword Only (no model)');
-  console.log('  3. Hybrid 60/40 (60% embedding, 40% keyword)');
-  console.log('  4. Hybrid 40/60 (40% embedding, 60% keyword)');
-  console.log('  5. Keyword-First (keywords primary, embedding tiebreaker)\n');
-
-  // RuvLTRA tests
-  console.log('─────────────────────────────────────────────────────────────────');
-  console.log('                       RUVLTRA RESULTS');
-  console.log('─────────────────────────────────────────────────────────────────\n');
-
-  const ruvEmbedding = runBenchmark(RUVLTRA_MODEL,
-    (task, taskEmb, agentEmbs) => routeEmbeddingOnly(taskEmb, agentEmbs),
-    'Embedding Only');
-  console.log(`  Embedding Only:   ${(ruvEmbedding.accuracy * 100).toFixed(1)}%`);
-
-  const ruvKeyword = runBenchmark(RUVLTRA_MODEL,
-    (task, taskEmb, agentEmbs) => routeKeywordOnly(task),
-    'Keyword Only');
-  console.log(`  Keyword Only:     ${(ruvKeyword.accuracy * 100).toFixed(1)}%`);
-
-  const ruvHybrid60 = runBenchmark(RUVLTRA_MODEL,
-    (task, taskEmb, agentEmbs) => routeHybrid(task, taskEmb, agentEmbs, 0.6, 0.4),
-    'Hybrid 60/40');
-  console.log(`  Hybrid 60/40:     ${(ruvHybrid60.accuracy * 100).toFixed(1)}%`);
-
-  const ruvHybrid40 = runBenchmark(RUVLTRA_MODEL,
-    (task, taskEmb, agentEmbs) => routeHybrid(task, taskEmb, agentEmbs, 0.4, 0.6),
-    'Hybrid 40/60');
-  console.log(`  Hybrid 40/60:     ${(ruvHybrid40.accuracy * 100).toFixed(1)}%`);
-
-  const ruvKwFirst = runBenchmark(RUVLTRA_MODEL,
-    (task, taskEmb, agentEmbs) => routeKeywordFirst(task, taskEmb, agentEmbs),
-    'Keyword-First');
-  console.log(`  Keyword-First:    ${(ruvKwFirst.accuracy * 100).toFixed(1)}%`);
-
-  // Qwen tests
-  console.log('\n─────────────────────────────────────────────────────────────────');
-  console.log('                       QWEN RESULTS');
-  console.log('─────────────────────────────────────────────────────────────────\n');
-
-  const qwenEmbedding = runBenchmark(QWEN_MODEL,
-    (task, taskEmb, agentEmbs) => routeEmbeddingOnly(taskEmb, agentEmbs),
-    'Embedding Only');
-  console.log(`  Embedding Only:   ${(qwenEmbedding.accuracy * 100).toFixed(1)}%`);
-
-  const qwenHybrid60 = runBenchmark(QWEN_MODEL,
-    (task, taskEmb, agentEmbs) => routeHybrid(task, taskEmb, agentEmbs, 0.6, 0.4),
-    'Hybrid 60/40');
-  console.log(`  Hybrid 60/40:     ${(qwenHybrid60.accuracy * 100).toFixed(1)}%`);
-
-  const qwenKwFirst = runBenchmark(QWEN_MODEL,
-    (task, taskEmb, agentEmbs) => routeKeywordFirst(task, taskEmb, agentEmbs),
-    'Keyword-First');
-  console.log(`  Keyword-First:    ${(qwenKwFirst.accuracy * 100).toFixed(1)}%`);
-
-  // Summary table
-  console.log('\n═══════════════════════════════════════════════════════════════════════════════════');
-  console.log('                              SUMMARY');
-  console.log('═══════════════════════════════════════════════════════════════════════════════════\n');
-
-  const fmt = (v) => `${(v * 100).toFixed(1)}%`.padStart(8);
-
-  console.log('┌───────────────────────┬──────────┬──────────┬──────────────────┐');
-  console.log('│ Strategy              │  RuvLTRA │   Qwen   │ RuvLTRA vs Qwen  │');
-  console.log('├───────────────────────┼──────────┼──────────┼──────────────────┤');
-  console.log(`│ Embedding Only        │${fmt(ruvEmbedding.accuracy)}  │${fmt(qwenEmbedding.accuracy)}  │     +${((ruvEmbedding.accuracy - qwenEmbedding.accuracy) * 100).toFixed(1)} pts     │`);
-  console.log(`│ Keyword Only          │${fmt(ruvKeyword.accuracy)}  │${fmt(ruvKeyword.accuracy)}  │       same       │`);
-  console.log(`│ Hybrid 60/40          │${fmt(ruvHybrid60.accuracy)}  │${fmt(qwenHybrid60.accuracy)}  │     +${((ruvHybrid60.accuracy - qwenHybrid60.accuracy) * 100).toFixed(1)} pts     │`);
-  console.log(`│ Keyword-First         │${fmt(ruvKwFirst.accuracy)}  │${fmt(qwenKwFirst.accuracy)}  │     +${((ruvKwFirst.accuracy - qwenKwFirst.accuracy) * 100).toFixed(1)} pts     │`);
-  console.log('└───────────────────────┴──────────┴──────────┴──────────────────┘');
-
-  // Best results
-  const ruvBest = [ruvEmbedding, ruvKeyword, ruvHybrid60, ruvHybrid40, ruvKwFirst]
-    .reduce((a, b) => a.accuracy > b.accuracy ? a : b);
-
-  console.log(`\n  BEST RuvLTRA: ${ruvBest.name} = ${(ruvBest.accuracy * 100).toFixed(1)}%`);
-  console.log(`  Improvement over embedding-only: +${((ruvBest.accuracy - ruvEmbedding.accuracy) * 100).toFixed(1)} points`);
-
-  // Show best results details
-  console.log('\n─────────────────────────────────────────────────────────────────');
-  console.log(`               BEST STRATEGY DETAILS: ${ruvBest.name}`);
-  console.log('─────────────────────────────────────────────────────────────────\n');
-
-  for (const r of ruvBest.results) {
-    const mark = r.correct ? '✓' : '✗';
-    const task = r.task.slice(0, 45).padEnd(45);
-    const exp = r.expected.padEnd(18);
-    console.log(`${mark} ${task} ${exp}${r.correct ? '' : '→ ' + r.got}`);
-  }
-
-  console.log('\n');
-}
-
-main().catch(console.error);
+#!/usr/bin/env node
+/**
+ * Hybrid Model Comparison
+ *
+ * Combines embedding similarity with keyword boosting.
+ * This addresses the "reviewer overfit" problem by:
+ * 1. Computing embedding similarity
+ * 2. Boosting agents that have keyword matches in the task
+ * 3. Using weighted combination for final score
+ */
+
+const { execSync } = require('child_process');
+const { existsSync } = require('fs');
+const { join } = require('path');
+const { homedir } = require('os');
+
+const MODELS_DIR = join(homedir(), '.ruvllm', 'models');
+const RUVLTRA_MODEL = join(MODELS_DIR, 'ruvltra-claude-code-0.5b-q4_k_m.gguf');
+const QWEN_MODEL = join(MODELS_DIR, 'qwen2.5-0.5b-instruct-q4_k_m.gguf');
+
+// V1 descriptions for embedding
+const DESCRIPTIONS_V1 = {
+  coder: 'implement create write build add code function class component feature',
+  researcher: 'research find investigate analyze explore search discover examine',
+  reviewer: 'review check evaluate assess inspect examine code quality',
+  tester: 'test unit integration e2e coverage mock assertion spec',
+  architect: 'design architecture schema system structure plan database',
+  'security-architect': 'security vulnerability xss injection audit cve authentication',
+  debugger: 'debug fix bug error issue broken crash exception trace',
+  documenter: 'document readme jsdoc comment explain describe documentation',
+  refactorer: 'refactor extract rename consolidate clean restructure simplify',
+  optimizer: 'optimize performance slow fast cache speed memory latency',
+  devops: 'deploy ci cd kubernetes docker pipeline container infrastructure',
+  'api-docs': 'openapi swagger api documentation graphql schema endpoint',
+  planner: 'plan estimate prioritize sprint roadmap schedule milestone',
+};
+
+// UNIQUE trigger keywords - words that strongly indicate a specific agent
+// Priority-ordered: first match wins for disambiguation
+// NOTE: "investigate" takes priority over "slow" for researcher vs optimizer
+const TRIGGER_KEYWORDS = {
+  // Higher priority agents (check these first)
+  researcher: ['research', 'investigate', 'explore', 'discover', 'best practices', 'patterns', 'analyze', 'look into', 'find out'],
+  coder: ['implement', 'build', 'create', 'component', 'function', 'typescript', 'react', 'feature', 'write code'],
+  tester: ['test', 'tests', 'testing', 'unit test', 'integration test', 'e2e', 'coverage', 'spec'],
+  reviewer: ['review', 'pull request', 'pr', 'code quality', 'code review', 'check code'],
+  debugger: ['debug', 'fix', 'bug', 'error', 'exception', 'crash', 'trace', 'null pointer', 'memory leak'],
+  'security-architect': ['security', 'vulnerability', 'xss', 'injection', 'csrf', 'cve', 'audit', 'exploit'],
+  refactorer: ['refactor', 'async/await', 'modernize', 'restructure', 'extract', 'legacy'],
+  // Optimizer: removed "slow" (too generic), added query-specific terms
+  optimizer: ['optimize', 'performance', 'cache', 'caching', 'speed up', 'latency', 'faster', 'queries', 'reduce time'],
+  architect: ['design', 'architecture', 'schema', 'structure', 'diagram', 'system design', 'plan architecture'],
+  documenter: ['jsdoc', 'comment', 'comments', 'readme', 'documentation', 'document', 'explain'],
+  devops: ['deploy', 'ci/cd', 'kubernetes', 'docker', 'pipeline', 'infrastructure', 'container'],
+  'api-docs': ['openapi', 'swagger', 'api doc', 'rest api', 'graphql', 'endpoint'],
+  planner: ['sprint', 'plan', 'roadmap', 'milestone', 'estimate', 'schedule', 'prioritize'],
+};
+
+// Priority order for disambiguation (when multiple agents match)
+const AGENT_PRIORITY = [
+  'researcher',     // "investigate" wins over "slow"
+  'debugger',       // "fix" wins over generic terms
+  'tester',         // "test" is specific
+  'security-architect',
+  'coder',
+  'reviewer',
+  'refactorer',
+  'optimizer',
+  'architect',
+  'documenter',
+  'devops',
+  'api-docs',
+  'planner',
+];
+
+const ROUTING_TESTS = [
+  { task: 'Implement a binary search function in TypeScript', expected: 'coder' },
+  { task: 'Write unit tests for the authentication module', expected: 'tester' },
+  { task: 'Review the pull request for security vulnerabilities', expected: 'reviewer' },
+  { task: 'Research best practices for React state management', expected: 'researcher' },
+  { task: 'Design the database schema for user profiles', expected: 'architect' },
+  { task: 'Fix the null pointer exception in the login handler', expected: 'debugger' },
+  { task: 'Audit the API endpoints for XSS vulnerabilities', expected: 'security-architect' },
+  { task: 'Write JSDoc comments for the utility functions', expected: 'documenter' },
+  { task: 'Refactor the payment module to use async/await', expected: 'refactorer' },
+  { task: 'Optimize the database queries for the dashboard', expected: 'optimizer' },
+  { task: 'Set up the CI/CD pipeline for the microservices', expected: 'devops' },
+  { task: 'Generate OpenAPI documentation for the REST API', expected: 'api-docs' },
+  { task: 'Create a sprint plan for the next two weeks', expected: 'planner' },
+  { task: 'Build a React component for user registration', expected: 'coder' },
+  { task: 'Debug memory leak in the WebSocket handler', expected: 'debugger' },
+  { task: 'Investigate slow API response times', expected: 'researcher' },
+  { task: 'Check code for potential race conditions', expected: 'reviewer' },
+  { task: 'Add integration tests for the payment gateway', expected: 'tester' },
+  { task: 'Plan the architecture for real-time notifications', expected: 'architect' },
+  { task: 'Cache the frequently accessed user data', expected: 'optimizer' },
+];
+
+function getEmbedding(modelPath, text) {
+  try {
+    const sanitized = text.replace(/"/g, '\\"').replace(/\n/g, ' ');
+    const result = execSync(
+      `llama-embedding -m "${modelPath}" -p "${sanitized}" --embd-output-format json 2>/dev/null`,
+      { encoding: 'utf-8', maxBuffer: 10 * 1024 * 1024 }
+    );
+    const json = JSON.parse(result);
+    return json.data[json.data.length - 1].embedding;
+  } catch {
+    return null;
+  }
+}
+
+function cosineSimilarity(a, b) {
+  if (!a || !b || a.length !== b.length) return 0;
+  let dot = 0, normA = 0, normB = 0;
+  for (let i = 0; i < a.length; i++) {
+    dot += a[i] * b[i];
+    normA += a[i] * a[i];
+    normB += b[i] * b[i];
+  }
+  return dot / (Math.sqrt(normA) * Math.sqrt(normB) || 1);
+}
+
+/**
+ * Count keyword matches for each agent
+ */
+function getKeywordScores(task) {
+  const taskLower = task.toLowerCase();
+  const scores = {};
+
+  for (const [agent, keywords] of Object.entries(TRIGGER_KEYWORDS)) {
+    let matches = 0;
+    for (const kw of keywords) {
+      if (taskLower.includes(kw.toLowerCase())) {
+        matches++;
+      }
+    }
+    scores[agent] = matches;
+  }
+
+  return scores;
+}
+
+/**
+ * Pure embedding routing (baseline)
+ */
+function routeEmbeddingOnly(taskEmbedding, agentEmbeddings) {
+  let bestAgent = 'coder';
+  let bestSim = -1;
+
+  for (const [agent, emb] of Object.entries(agentEmbeddings)) {
+    const sim = cosineSimilarity(taskEmbedding, emb);
+    if (sim > bestSim) {
+      bestSim = sim;
+      bestAgent = agent;
+    }
+  }
+
+  return { agent: bestAgent, confidence: bestSim };
+}
+
+/**
+ * Pure keyword routing
+ */
+function routeKeywordOnly(task) {
+  const scores = getKeywordScores(task);
+  let bestAgent = 'coder';
+  let bestScore = 0;
+
+  for (const [agent, score] of Object.entries(scores)) {
+    if (score > bestScore) {
+      bestScore = score;
+      bestAgent = agent;
+    }
+  }
+
+  return { agent: bestAgent, confidence: bestScore };
+}
+
+/**
+ * Hybrid routing - combine embedding similarity with keyword boost
+ */
+function routeHybrid(task, taskEmbedding, agentEmbeddings, embeddingWeight = 0.6, keywordWeight = 0.4) {
+  const keywordScores = getKeywordScores(task);
+
+  // Normalize keyword scores to 0-1 range
+  const maxKeyword = Math.max(...Object.values(keywordScores), 1);
+  const normalizedKeywords = {};
+  for (const agent of Object.keys(keywordScores)) {
+    normalizedKeywords[agent] = keywordScores[agent] / maxKeyword;
+  }
+
+  let bestAgent = 'coder';
+  let bestScore = -1;
+  const allScores = {};
+
+  for (const [agent, emb] of Object.entries(agentEmbeddings)) {
+    const embSim = cosineSimilarity(taskEmbedding, emb);
+    const kwScore = normalizedKeywords[agent] || 0;
+    const combined = embeddingWeight * embSim + keywordWeight * kwScore;
+    allScores[agent] = { embedding: embSim, keyword: kwScore, combined };
+
+    if (combined > bestScore) {
+      bestScore = combined;
+      bestAgent = agent;
+    }
+  }
+
+  return { agent: bestAgent, confidence: bestScore, scores: allScores };
+}
+
+/**
+ * Keyword-first routing - use keywords as primary, embedding as tiebreaker
+ */
+function routeKeywordFirst(task, taskEmbedding, agentEmbeddings) {
+  const keywordScores = getKeywordScores(task);
+
+  // Find agents with max keyword matches
+  const maxKw = Math.max(...Object.values(keywordScores));
+
+  if (maxKw > 0) {
+    // At least one keyword match - use keywords, embedding as tiebreaker
+    const candidates = Object.entries(keywordScores)
+      .filter(([_, score]) => score === maxKw)
+      .map(([agent, _]) => agent);
+
+    if (candidates.length === 1) {
+      return { agent: candidates[0], confidence: maxKw };
+    }
+
+    // Multiple candidates with same keyword count - use embedding
+    let bestAgent = candidates[0];
+    let bestSim = -1;
+    for (const agent of candidates) {
+      const sim = cosineSimilarity(taskEmbedding, agentEmbeddings[agent]);
+      if (sim > bestSim) {
+        bestSim = sim;
+        bestAgent = agent;
+      }
+    }
+    return { agent: bestAgent, confidence: maxKw + bestSim / 10 };
+  }
+
+  // No keyword matches - fall back to pure embedding
+  return routeEmbeddingOnly(taskEmbedding, agentEmbeddings);
+}
+
+function runBenchmark(modelPath, routerFn, name) {
+  const agentEmbeddings = {};
+  for (const [agent, desc] of Object.entries(DESCRIPTIONS_V1)) {
+    agentEmbeddings[agent] = getEmbedding(modelPath, desc);
+  }
+
+  let correct = 0;
+  const results = [];
+
+  for (const test of ROUTING_TESTS) {
+    const taskEmb = getEmbedding(modelPath, test.task);
+    const { agent } = routerFn(test.task, taskEmb, agentEmbeddings);
+    const isCorrect = agent === test.expected;
+    if (isCorrect) correct++;
+    results.push({ task: test.task, expected: test.expected, got: agent, correct: isCorrect });
+  }
+
+  return { accuracy: correct / ROUTING_TESTS.length, correct, total: ROUTING_TESTS.length, results, name };
+}
+
+async function main() {
+  console.log('\n╔═══════════════════════════════════════════════════════════════════════════════════╗');
+  console.log('║                 HYBRID ROUTING: Embeddings + Keywords                             ║');
+  console.log('╚═══════════════════════════════════════════════════════════════════════════════════╝\n');
+
+  if (!existsSync(RUVLTRA_MODEL)) {
+    console.error('RuvLTRA model not found.');
+    process.exit(1);
+  }
+
+  console.log('Strategies:');
+  console.log('  1. Embedding Only (baseline)');
+  console.log('  2. Keyword Only (no model)');
+  console.log('  3. Hybrid 60/40 (60% embedding, 40% keyword)');
+  console.log('  4. Hybrid 40/60 (40% embedding, 60% keyword)');
+  console.log('  5. Keyword-First (keywords primary, embedding tiebreaker)\n');
+
+  // RuvLTRA tests
+  console.log('─────────────────────────────────────────────────────────────────');
+  console.log('                       RUVLTRA RESULTS');
+  console.log('─────────────────────────────────────────────────────────────────\n');
+
+  const ruvEmbedding = runBenchmark(RUVLTRA_MODEL,
+    (task, taskEmb, agentEmbs) => routeEmbeddingOnly(taskEmb, agentEmbs),
+    'Embedding Only');
+  console.log(`  Embedding Only:   ${(ruvEmbedding.accuracy * 100).toFixed(1)}%`);
+
+  const ruvKeyword = runBenchmark(RUVLTRA_MODEL,
+    (task, taskEmb, agentEmbs) => routeKeywordOnly(task),
+    'Keyword Only');
+  console.log(`  Keyword Only:     ${(ruvKeyword.accuracy * 100).toFixed(1)}%`);
+
+  const ruvHybrid60 = runBenchmark(RUVLTRA_MODEL,
+    (task, taskEmb, agentEmbs) => routeHybrid(task, taskEmb, agentEmbs, 0.6, 0.4),
+    'Hybrid 60/40');
+  console.log(`  Hybrid 60/40:     ${(ruvHybrid60.accuracy * 100).toFixed(1)}%`);
+
+  const ruvHybrid40 = runBenchmark(RUVLTRA_MODEL,
+    (task, taskEmb, agentEmbs) => routeHybrid(task, taskEmb, agentEmbs, 0.4, 0.6),
+    'Hybrid 40/60');
+  console.log(`  Hybrid 40/60:     ${(ruvHybrid40.accuracy * 100).toFixed(1)}%`);
+
+  const ruvKwFirst = runBenchmark(RUVLTRA_MODEL,
+    (task, taskEmb, agentEmbs) => routeKeywordFirst(task, taskEmb, agentEmbs),
+    'Keyword-First');
+  console.log(`  Keyword-First:    ${(ruvKwFirst.accuracy * 100).toFixed(1)}%`);
+
+  // Qwen tests
+  console.log('\n─────────────────────────────────────────────────────────────────');
+  console.log('                       QWEN RESULTS');
+  console.log('─────────────────────────────────────────────────────────────────\n');
+
+  const qwenEmbedding = runBenchmark(QWEN_MODEL,
+    (task, taskEmb, agentEmbs) => routeEmbeddingOnly(taskEmb, agentEmbs),
+    'Embedding Only');
+  console.log(`  Embedding Only:   ${(qwenEmbedding.accuracy * 100).toFixed(1)}%`);
+
+  const qwenHybrid60 = runBenchmark(QWEN_MODEL,
+    (task, taskEmb, agentEmbs) => routeHybrid(task, taskEmb, agentEmbs, 0.6, 0.4),
+    'Hybrid 60/40');
+  console.log(`  Hybrid 60/40:     ${(qwenHybrid60.accuracy * 100).toFixed(1)}%`);
+
+  const qwenKwFirst = runBenchmark(QWEN_MODEL,
+    (task, taskEmb, agentEmbs) => routeKeywordFirst(task, taskEmb, agentEmbs),
+    'Keyword-First');
+  console.log(`  Keyword-First:    ${(qwenKwFirst.accuracy * 100).toFixed(1)}%`);
+
+  // Summary table
+  console.log('\n═══════════════════════════════════════════════════════════════════════════════════');
+  console.log('                              SUMMARY');
+  console.log('═══════════════════════════════════════════════════════════════════════════════════\n');
+
+  const fmt = (v) => `${(v * 100).toFixed(1)}%`.padStart(8);
+
+  console.log('┌───────────────────────┬──────────┬──────────┬──────────────────┐');
+  console.log('│ Strategy              │  RuvLTRA │   Qwen   │ RuvLTRA vs Qwen  │');
+  console.log('├───────────────────────┼──────────┼──────────┼──────────────────┤');
+  console.log(`│ Embedding Only        │${fmt(ruvEmbedding.accuracy)}  │${fmt(qwenEmbedding.accuracy)}  │     +${((ruvEmbedding.accuracy - qwenEmbedding.accuracy) * 100).toFixed(1)} pts     │`);
+  console.log(`│ Keyword Only          │${fmt(ruvKeyword.accuracy)}  │${fmt(ruvKeyword.accuracy)}  │       same       │`);
+  console.log(`│ Hybrid 60/40          │${fmt(ruvHybrid60.accuracy)}  │${fmt(qwenHybrid60.accuracy)}  │     +${((ruvHybrid60.accuracy - qwenHybrid60.accuracy) * 100).toFixed(1)} pts     │`);
+  console.log(`│ Keyword-First         │${fmt(ruvKwFirst.accuracy)}  │${fmt(qwenKwFirst.accuracy)}  │     +${((ruvKwFirst.accuracy - qwenKwFirst.accuracy) * 100).toFixed(1)} pts     │`);
+  console.log('└───────────────────────┴──────────┴──────────┴──────────────────┘');
+
+  // Best results
+  const ruvBest = [ruvEmbedding, ruvKeyword, ruvHybrid60, ruvHybrid40, ruvKwFirst]
+    .reduce((a, b) => a.accuracy > b.accuracy ? a : b);
+
+  console.log(`\n  BEST RuvLTRA: ${ruvBest.name} = ${(ruvBest.accuracy * 100).toFixed(1)}%`);
+  console.log(`  Improvement over embedding-only: +${((ruvBest.accuracy - ruvEmbedding.accuracy) * 100).toFixed(1)} points`);
+
+  // Show best results details
+  console.log('\n─────────────────────────────────────────────────────────────────');
+  console.log(`               BEST STRATEGY DETAILS: ${ruvBest.name}`);
+  console.log('─────────────────────────────────────────────────────────────────\n');
+
+  for (const r of ruvBest.results) {
+    const mark = r.correct ? '✓' : '✗';
+    const task = r.task.slice(0, 45).padEnd(45);
+    const exp = r.expected.padEnd(18);
+    console.log(`${mark} ${task} ${exp}${r.correct ? '' : '→ ' + r.got}`);
+  }
+
+  console.log('\n');
+}
+
+main().catch(console.error);