ai-functions 0.3.0 → 0.4.0

package/test/e2e-bedrock-manual.ts

@@ -0,0 +1,163 @@
+#!/usr/bin/env npx tsx
+/**
+ * Manual E2E Test for AWS Bedrock Flex Tier Processing
+ *
+ * Run with your AWS credentials:
+ *
+ *   AWS_ACCESS_KEY_ID=... AWS_SECRET_ACCESS_KEY=... npx tsx test/e2e-bedrock-manual.ts
+ *
+ * Or if using AWS SSO/profiles:
+ *
+ *   AWS_PROFILE=your-profile npx tsx test/e2e-bedrock-manual.ts
+ */
+
+import { configure, resetContext, getExecutionTier, isFlexAvailable, getProvider } from '../src/context.js'
+import { type BatchItem } from '../src/batch-queue.js'
+import { configureAWSBedrock, bedrockFlexAdapter } from '../src/batch/bedrock.js'
+
+// Model IDs on Bedrock
+const CLAUDE_OPUS_MODEL = 'anthropic.claude-opus-4-20250514-v1:0'
+const CLAUDE_SONNET_MODEL = 'anthropic.claude-sonnet-4-20250514-v1:0'
+
+async function main() {
+  console.log('\n🧪 E2E AWS Bedrock Flex Tier Test\n')
+
+  // Check for AWS credentials
+  if (!process.env.AWS_ACCESS_KEY_ID && !process.env.AWS_PROFILE) {
+    console.error('❌ AWS credentials required')
+    console.error('   Set AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY')
+    console.error('   Or use AWS_PROFILE for SSO/credential profiles')
+    process.exit(1)
+  }
+
+  console.log('✅ AWS credentials found')
+  console.log(`   Region: ${process.env.AWS_REGION || 'us-east-1'}`)
+
+  // Configure Bedrock
+  configureAWSBedrock({
+    region: process.env.AWS_REGION || 'us-east-1',
+    s3Bucket: 'dummy-bucket', // Flex mode doesn't use S3
+  })
+
+  configure({
+    provider: 'bedrock',
+    model: CLAUDE_SONNET_MODEL,
+    batchMode: 'auto',
+    flexThreshold: 3,
+    batchThreshold: 500,
+  })
+
+  console.log(`✅ Configured: provider=${getProvider()}, flexAvailable=${isFlexAvailable()}`)
+
+  // Test 1: Basic flex processing with Claude Sonnet
+  console.log('\n📝 Test 1: Basic Flex Processing (Claude Sonnet)')
+  console.log('   Creating 3 simple prompts...')
+
+  const items: BatchItem[] = [
+    { id: 'test_1', prompt: 'What is 2 + 2? Reply with just the number.', status: 'pending' },
+    { id: 'test_2', prompt: 'What is 3 + 3? Reply with just the number.', status: 'pending' },
+    { id: 'test_3', prompt: 'What is 4 + 4? Reply with just the number.', status: 'pending' },
+  ]
+
+  console.log('   Submitting via flex adapter...')
+  const startTime = Date.now()
+
+  try {
+    const results = await bedrockFlexAdapter.submitFlex(items, { model: CLAUDE_SONNET_MODEL })
+    const duration = Date.now() - startTime
+
+    console.log(`   ✅ Completed in ${duration}ms`)
+    console.log('   Results:')
+
+    results.forEach((result) => {
+      const status = result.status === 'completed' ? '✅' : '❌'
+      console.log(`     ${status} ${result.id}: "${result.result}"`)
+      if (result.error) {
+        console.log(`        Error: ${result.error}`)
+      }
+    })
+  } catch (error) {
+    console.error(`   ❌ Error: ${error}`)
+    process.exit(1)
+  }
+
+  // Test 2: Token usage
+  console.log('\n📝 Test 2: Token Usage Tracking')
+
+  const usageItems: BatchItem[] = [
+    { id: 'usage', prompt: 'Count from 1 to 5', status: 'pending' },
+  ]
+
+  try {
+    const results = await bedrockFlexAdapter.submitFlex(usageItems, { model: CLAUDE_SONNET_MODEL })
+    const usage = results[0].usage
+
+    if (usage) {
+      console.log(`   Prompt tokens: ${usage.promptTokens}`)
+      console.log(`   Completion tokens: ${usage.completionTokens}`)
+      console.log(`   Total tokens: ${usage.totalTokens}`)
+      console.log('   ✅ Token usage tracking works')
+    } else {
+      console.log('   ⚠️  No usage data returned')
+    }
+  } catch (error) {
+    console.error(`   ❌ Error: ${error}`)
+  }
+
+  // Test 3: Try Claude Opus 4.5 (may not be available)
+  console.log('\n📝 Test 3: Claude Opus 4.5 (optional)')
+
+  const opusItems: BatchItem[] = [
+    { id: 'opus', prompt: 'Write a haiku about AI.', status: 'pending' },
+  ]
+
+  try {
+    const results = await bedrockFlexAdapter.submitFlex(opusItems, { model: CLAUDE_OPUS_MODEL })
+    console.log(`   ✅ Claude Opus 4.5 response:`)
+    console.log(`   ${results[0].result}`)
+  } catch (error) {
+    console.log(`   ⚠️  Claude Opus 4.5 not available: ${error}`)
+    console.log('   (This is expected if Opus is not enabled in your AWS account)')
+  }
+
+  // Test 4: Concurrent processing
+  console.log('\n📝 Test 4: Concurrent Processing (5 items)')
+
+  const concurrentItems: BatchItem[] = Array.from({ length: 5 }, (_, i) => ({
+    id: `concurrent_${i}`,
+    prompt: `What is ${i + 1} * 2? Reply with just the number.`,
+    status: 'pending' as const,
+  }))
+
+  try {
+    const startTime = Date.now()
+    const results = await bedrockFlexAdapter.submitFlex(concurrentItems, { model: CLAUDE_SONNET_MODEL })
+    const duration = Date.now() - startTime
+
+    const successCount = results.filter((r) => r.status === 'completed').length
+    console.log(`   ✅ Processed ${successCount}/5 items in ${duration}ms`)
+
+    results.forEach((result) => {
+      console.log(`     ${result.id}: "${result.result}"`)
+    })
+  } catch (error) {
+    console.error(`   ❌ Error: ${error}`)
+  }
+
+  // Test 5: Tier selection
+  console.log('\n📝 Test 5: Execution Tier Selection')
+
+  console.log(`   1 item   → ${getExecutionTier(1)} (expected: immediate)`)
+  console.log(`   3 items  → ${getExecutionTier(3)} (expected: flex)`)
+  console.log(`   10 items → ${getExecutionTier(10)} (expected: flex)`)
+  console.log(`   500 items → ${getExecutionTier(500)} (expected: batch)`)
+
+  console.log('\n✨ E2E Tests Complete!\n')
+
+  resetContext()
+}
+
+main().catch((error) => {
+  console.error('Fatal error:', error)
+  process.exit(1)
+})

package/test/e2e-bedrock.test.ts

@@ -0,0 +1,191 @@
+/**
+ * E2E Tests for AWS Bedrock Flex Tier Processing
+ *
+ * These tests hit real AWS Bedrock APIs to verify the flex tier actually works.
+ * Requires AWS credentials:
+ * - AWS_ACCESS_KEY_ID
+ * - AWS_SECRET_ACCESS_KEY
+ * - AWS_REGION (optional, defaults to us-east-1)
+ *
+ * Run with: npx vitest run test/e2e-bedrock.test.ts
+ */
+
+import { describe, it, expect, beforeAll, afterAll } from 'vitest'
+import { configure, resetContext, getExecutionTier, isFlexAvailable } from '../src/context.js'
+import { getFlexAdapter, type BatchItem } from '../src/batch-queue.js'
+
+// Import the Bedrock adapter to register it
+import { configureAWSBedrock, bedrockFlexAdapter } from '../src/batch/bedrock.js'
+import '../src/batch/bedrock.js'
+
+// Skip tests if no AWS credentials
+// NOTE: Unlike OpenAI/Google, Bedrock via AI Gateway still requires AWS SigV4 signing
+// Gateway alone is NOT sufficient - you need AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY
+const hasCredentials = !!(process.env.AWS_ACCESS_KEY_ID && process.env.AWS_SECRET_ACCESS_KEY)
+
+// Claude Opus 4.5 model ID on Bedrock
+const CLAUDE_OPUS_MODEL = 'anthropic.claude-opus-4-20250514-v1:0'
+// Fallback to Sonnet if Opus isn't available
+const CLAUDE_SONNET_MODEL = 'anthropic.claude-sonnet-4-20250514-v1:0'
+
+describe.skipIf(!hasCredentials)('E2E Bedrock Flex Tier Processing', () => {
+  beforeAll(() => {
+    // Configure Bedrock - can use either direct AWS credentials or AI Gateway
+    configureAWSBedrock({
+      region: process.env.AWS_REGION || 'us-east-1',
+      s3Bucket: process.env.BEDROCK_BATCH_S3_BUCKET || 'dummy-bucket',
+      // AI Gateway configuration (optional)
+      gatewayUrl: process.env.AI_GATEWAY_URL,
+      gatewayToken: process.env.AI_GATEWAY_TOKEN,
+    })
+
+    configure({
+      provider: 'bedrock',
+      model: CLAUDE_SONNET_MODEL, // Use Sonnet by default (more available)
+      batchMode: 'auto',
+      flexThreshold: 3,
+      batchThreshold: 500,
+    })
+  })
+
+  afterAll(() => {
+    resetContext()
+  })
+
+  describe('Flex Adapter Direct', () => {
+    it('processes a small batch via flex adapter', async () => {
+      const items: BatchItem[] = [
+        {
+          id: 'item_1',
+          prompt: 'What is 2 + 2? Reply with just the number.',
+          status: 'pending',
+        },
+        {
+          id: 'item_2',
+          prompt: 'What is 3 + 3? Reply with just the number.',
+          status: 'pending',
+        },
+        {
+          id: 'item_3',
+          prompt: 'What is 4 + 4? Reply with just the number.',
+          status: 'pending',
+        },
+      ]
+
+      const results = await bedrockFlexAdapter.submitFlex(items, { model: CLAUDE_SONNET_MODEL })
+
+      expect(results).toHaveLength(3)
+      results.forEach((result) => {
+        expect(result.status).toBe('completed')
+        expect(result.result).toBeDefined()
+      })
+
+      // Check the actual responses contain numbers
+      const result1 = results.find((r) => r.id === 'item_1')
+      const result2 = results.find((r) => r.id === 'item_2')
+      const result3 = results.find((r) => r.id === 'item_3')
+
+      expect(String(result1?.result)).toContain('4')
+      expect(String(result2?.result)).toContain('6')
+      expect(String(result3?.result)).toContain('8')
+    }, 60000) // 60 second timeout for Bedrock
+
+    it('reports token usage', async () => {
+      const items: BatchItem[] = [
+        {
+          id: 'usage_test',
+          prompt: 'Count from 1 to 5',
+          status: 'pending',
+        },
+      ]
+
+      const results = await bedrockFlexAdapter.submitFlex(items, { model: CLAUDE_SONNET_MODEL })
+
+      expect(results[0].usage).toBeDefined()
+      expect(results[0].usage?.promptTokens).toBeGreaterThan(0)
+      expect(results[0].usage?.completionTokens).toBeGreaterThan(0)
+      expect(results[0].usage?.totalTokens).toBeGreaterThan(0)
+    }, 60000)
+
+    it('handles system prompts', async () => {
+      const items: BatchItem[] = [
+        {
+          id: 'system_test',
+          prompt: 'What should I say?',
+          options: {
+            system: 'You are a helpful assistant that always responds with exactly "Hello World"',
+          },
+          status: 'pending',
+        },
+      ]
+
+      const results = await bedrockFlexAdapter.submitFlex(items, { model: CLAUDE_SONNET_MODEL })
+
+      expect(results[0].status).toBe('completed')
+      expect(String(results[0].result).toLowerCase()).toContain('hello')
+    }, 60000)
+  })
+
+  describe('Execution Tier Selection', () => {
+    it('selects immediate tier for < flexThreshold items', () => {
+      configure({ batchMode: 'auto', flexThreshold: 5, batchThreshold: 500 })
+      expect(getExecutionTier(1)).toBe('immediate')
+      expect(getExecutionTier(4)).toBe('immediate')
+    })
+
+    it('selects flex tier for flexThreshold to < batchThreshold items', () => {
+      configure({ batchMode: 'auto', flexThreshold: 5, batchThreshold: 500 })
+      expect(getExecutionTier(5)).toBe('flex')
+      expect(getExecutionTier(100)).toBe('flex')
+    })
+
+    it('reports flex as available for bedrock', () => {
+      configure({ provider: 'bedrock' })
+      expect(isFlexAvailable()).toBe(true)
+    })
+  })
+
+  describe('Concurrent Processing', () => {
+    it('processes multiple items concurrently', async () => {
+      // Create 5 items to test concurrent processing
+      const items: BatchItem[] = Array.from({ length: 5 }, (_, i) => ({
+        id: `concurrent_${i}`,
+        prompt: `What is ${i} + 1? Reply with just the number.`,
+        status: 'pending' as const,
+      }))
+
+      const startTime = Date.now()
+      const results = await bedrockFlexAdapter.submitFlex(items, { model: CLAUDE_SONNET_MODEL })
+      const duration = Date.now() - startTime
+
+      expect(results).toHaveLength(5)
+      results.forEach((result) => {
+        expect(result.status).toBe('completed')
+      })
+
+      console.log(`Processed 5 items in ${duration}ms`)
+    }, 120000)
+  })
+})
+
+// Optional test with Claude Opus 4.5 (may not be available in all regions)
+describe.skipIf(!hasCredentials)('E2E Bedrock with Claude Opus 4.5', () => {
+  it.skip('processes with Claude Opus 4.5', async () => {
+    const items: BatchItem[] = [
+      {
+        id: 'opus_test',
+        prompt: 'Write a haiku about AI.',
+        status: 'pending',
+      },
+    ]
+
+    try {
+      const results = await bedrockFlexAdapter.submitFlex(items, { model: CLAUDE_OPUS_MODEL })
+      expect(results[0].status).toBe('completed')
+      expect(results[0].result).toBeDefined()
+      console.log('Claude Opus 4.5 response:', results[0].result)
+    } catch (error) {
+      console.log('Claude Opus 4.5 not available in this region:', error)
+    }
+  }, 120000)
+})

package/test/e2e-flex-gateway.ts

@@ -0,0 +1,157 @@
+#!/usr/bin/env npx tsx
+/**
+ * E2E Test for Flex Tier via Cloudflare AI Gateway
+ *
+ * This test uses the Cloudflare AI Gateway to route requests to OpenAI,
+ * using the AI_GATEWAY_TOKEN and AI_GATEWAY_URL environment variables.
+ *
+ * Run with:
+ *   npx tsx test/e2e-flex-gateway.ts
+ */
+
+import { configure, resetContext, getExecutionTier, isFlexAvailable, getProvider } from '../src/context.js'
+import { getFlexAdapter, type BatchItem } from '../src/batch-queue.js'
+import { configureOpenAI, openaiFlexAdapter } from '../src/batch/openai.js'
+
+async function main() {
+  console.log('\n🧪 E2E Flex Tier Test (via Cloudflare AI Gateway)\n')
+
+  // Check for gateway config
+  const gatewayUrl = process.env.AI_GATEWAY_URL
+  const gatewayToken = process.env.AI_GATEWAY_TOKEN
+
+  if (!gatewayUrl || !gatewayToken) {
+    console.error('❌ AI_GATEWAY_URL and AI_GATEWAY_TOKEN environment variables are required')
+    console.error('   These should be in .env at the repo root')
+    process.exit(1)
+  }
+
+  console.log(`✅ Gateway URL: ${gatewayUrl}`)
+  console.log('✅ Gateway Token found')
+
+  // Configure OpenAI adapter to use the gateway
+  // The gateway URL format is: https://gateway.ai.cloudflare.com/v1/{account_id}/{gateway_id}
+  // We need to append /openai to route to OpenAI
+  const openaiGatewayUrl = `${gatewayUrl}/openai`
+  console.log(`   Using base URL: ${openaiGatewayUrl}`)
+
+  configureOpenAI({
+    apiKey: gatewayToken,
+    baseUrl: openaiGatewayUrl,
+  })
+
+  // Configure context
+  configure({
+    provider: 'openai',
+    model: 'gpt-4o-mini',
+    batchMode: 'auto',
+    flexThreshold: 3,
+    batchThreshold: 500,
+  })
+
+  console.log(`✅ Configured: provider=${getProvider()}, flexAvailable=${isFlexAvailable()}`)
+
+  // Test 1: Basic flex processing
+  console.log('\n📝 Test 1: Basic Flex Processing')
+  console.log('   Creating 3 simple prompts...')
+
+  const items: BatchItem[] = [
+    { id: 'test_1', prompt: 'Say "hello" and nothing else.', status: 'pending' },
+    { id: 'test_2', prompt: 'Say "world" and nothing else.', status: 'pending' },
+    { id: 'test_3', prompt: 'Say "test" and nothing else.', status: 'pending' },
+  ]
+
+  console.log('   Submitting via flex adapter...')
+  const startTime = Date.now()
+
+  try {
+    const results = await openaiFlexAdapter.submitFlex(items, { model: 'gpt-4o-mini' })
+    const duration = Date.now() - startTime
+
+    console.log(`   ✅ Completed in ${duration}ms`)
+    console.log('   Results:')
+
+    results.forEach((result) => {
+      const status = result.status === 'completed' ? '✅' : '❌'
+      console.log(`     ${status} ${result.id}: "${result.result}"`)
+      if (result.error) {
+        console.log(`        Error: ${result.error}`)
+      }
+    })
+
+    const successCount = results.filter((r) => r.status === 'completed').length
+    if (successCount === results.length) {
+      console.log(`   ✅ All ${successCount} items completed successfully`)
+    } else {
+      console.log(`   ⚠️  ${successCount}/${results.length} items completed`)
+    }
+  } catch (error) {
+    console.error(`   ❌ Error: ${error}`)
+    process.exit(1)
+  }
+
+  // Test 2: Token usage
+  console.log('\n📝 Test 2: Token Usage Tracking')
+
+  const usageItems: BatchItem[] = [
+    { id: 'usage', prompt: 'Count from 1 to 5', status: 'pending' },
+  ]
+
+  try {
+    const results = await openaiFlexAdapter.submitFlex(usageItems, { model: 'gpt-4o-mini' })
+    const usage = results[0].usage
+
+    if (usage) {
+      console.log(`   Prompt tokens: ${usage.promptTokens}`)
+      console.log(`   Completion tokens: ${usage.completionTokens}`)
+      console.log(`   Total tokens: ${usage.totalTokens}`)
+      console.log('   ✅ Token usage tracking works')
+    } else {
+      console.log('   ⚠️  No usage data returned')
+    }
+  } catch (error) {
+    console.error(`   ❌ Error: ${error}`)
+  }
+
+  // Test 3: Concurrent processing
+  console.log('\n📝 Test 3: Concurrent Processing (5 items)')
+
+  const concurrentItems: BatchItem[] = Array.from({ length: 5 }, (_, i) => ({
+    id: `concurrent_${i}`,
+    prompt: `What is ${i + 1} + ${i + 1}? Reply with just the number.`,
+    status: 'pending' as const,
+  }))
+
+  try {
+    const startTime = Date.now()
+    const results = await openaiFlexAdapter.submitFlex(concurrentItems, { model: 'gpt-4o-mini' })
+    const duration = Date.now() - startTime
+
+    const successCount = results.filter((r) => r.status === 'completed').length
+    console.log(`   ✅ Processed ${successCount}/5 items in ${duration}ms`)
+    console.log(`   Average: ${Math.round(duration / 5)}ms per item`)
+
+    results.forEach((result) => {
+      console.log(`     ${result.id}: "${result.result}"`)
+    })
+  } catch (error) {
+    console.error(`   ❌ Error: ${error}`)
+  }
+
+  // Test 4: Tier selection
+  console.log('\n📝 Test 4: Execution Tier Selection')
+
+  console.log(`   1 item   → ${getExecutionTier(1)} (expected: immediate)`)
+  console.log(`   3 items  → ${getExecutionTier(3)} (expected: flex)`)
+  console.log(`   10 items → ${getExecutionTier(10)} (expected: flex)`)
+  console.log(`   500 items → ${getExecutionTier(500)} (expected: batch)`)
+
+  console.log('\n✨ E2E Tests Complete!\n')
+
+  resetContext()
+}
+
+main().catch((error) => {
+  console.error('Fatal error:', error)
+  process.exit(1)
+})

package/test/e2e-flex-manual.ts

@@ -0,0 +1,183 @@
+#!/usr/bin/env npx tsx
+/**
+ * Manual E2E Test for Flex Tier Processing
+ *
+ * Run this script manually with your OpenAI API key:
+ *
+ *   OPENAI_API_KEY=sk-... npx tsx test/e2e-flex-manual.ts
+ *
+ * Or if you have the key in your environment:
+ *
+ *   npx tsx test/e2e-flex-manual.ts
+ */
+
+import { configure, resetContext, getExecutionTier, isFlexAvailable, getProvider } from '../src/context.js'
+import { getFlexAdapter, type BatchItem } from '../src/batch-queue.js'
+
+// Import the OpenAI adapter to register it
+import '../src/batch/openai.js'
+
+async function main() {
+  console.log('\n🧪 E2E Flex Tier Test\n')
+
+  // Check for API key
+  if (!process.env.OPENAI_API_KEY) {
+    console.error('❌ OPENAI_API_KEY environment variable is required')
+    console.error('   Run with: OPENAI_API_KEY=sk-... npx tsx test/e2e-flex-manual.ts')
+    process.exit(1)
+  }
+
+  console.log('✅ OpenAI API key found')
+
+  // Configure
+  configure({
+    provider: 'openai',
+    model: 'gpt-4o-mini',
+    batchMode: 'auto',
+    flexThreshold: 3,
+    batchThreshold: 500,
+  })
+
+  console.log(`✅ Configured: provider=${getProvider()}, flexAvailable=${isFlexAvailable()}`)
+
+  // Test 1: Basic flex processing
+  console.log('\n📝 Test 1: Basic Flex Processing')
+  console.log('   Creating 5 simple math prompts...')
+
+  const adapter = getFlexAdapter('openai')
+
+  const items: BatchItem[] = [
+    { id: 'math_1', prompt: 'What is 2 + 2? Reply with just the number.', status: 'pending' },
+    { id: 'math_2', prompt: 'What is 3 + 3? Reply with just the number.', status: 'pending' },
+    { id: 'math_3', prompt: 'What is 4 + 4? Reply with just the number.', status: 'pending' },
+    { id: 'math_4', prompt: 'What is 5 + 5? Reply with just the number.', status: 'pending' },
+    { id: 'math_5', prompt: 'What is 6 + 6? Reply with just the number.', status: 'pending' },
+  ]
+
+  console.log('   Submitting via flex adapter...')
+  const startTime = Date.now()
+
+  try {
+    const results = await adapter.submitFlex(items, { model: 'gpt-4o-mini' })
+    const duration = Date.now() - startTime
+
+    console.log(`   ✅ Completed in ${duration}ms`)
+    console.log('   Results:')
+
+    let allPassed = true
+    const expected = ['4', '6', '8', '10', '12']
+
+    results.forEach((result, i) => {
+      const passed = result.status === 'completed' && result.result?.toString().includes(expected[i])
+      const icon = passed ? '✅' : '❌'
+      console.log(`     ${icon} ${result.id}: "${result.result}" (expected: ${expected[i]})`)
+      if (!passed) allPassed = false
+    })
+
+    if (!allPassed) {
+      console.log('\n⚠️  Some results did not match expected values')
+    }
+  } catch (error) {
+    console.error(`   ❌ Error: ${error}`)
+    process.exit(1)
+  }
+
+  // Test 2: Structured output
+  console.log('\n📝 Test 2: Structured Output (JSON Schema)')
+
+  const structuredItems: BatchItem[] = [
+    {
+      id: 'person_1',
+      prompt: 'Generate a JSON object with name "Alice" and age 30. Format: {"name": "...", "age": ...}',
+      schema: { name: 'string', age: 'number' },
+      status: 'pending',
+    },
+    {
+      id: 'person_2',
+      prompt: 'Generate a JSON object with name "Bob" and age 25. Format: {"name": "...", "age": ...}',
+      schema: { name: 'string', age: 'number' },
+      status: 'pending',
+    },
+  ]
+
+  try {
+    const results = await adapter.submitFlex(structuredItems, { model: 'gpt-4o-mini' })
+
+    console.log('   Results:')
+    results.forEach((result) => {
+      const obj = result.result as { name?: string; age?: number }
+      console.log(`     ${result.id}: name="${obj?.name}", age=${obj?.age}`)
+    })
+    console.log('   ✅ Structured output works')
+  } catch (error) {
+    console.error(`   ❌ Error: ${error}`)
+  }
+
+  // Test 3: Token usage
+  console.log('\n📝 Test 3: Token Usage Tracking')
+
+  const usageItems: BatchItem[] = [
+    { id: 'usage', prompt: 'Count from 1 to 10', status: 'pending' },
+  ]
+
+  try {
+    const results = await adapter.submitFlex(usageItems, { model: 'gpt-4o-mini' })
+    const usage = results[0].usage
+
+    if (usage) {
+      console.log(`   Prompt tokens: ${usage.promptTokens}`)
+      console.log(`   Completion tokens: ${usage.completionTokens}`)
+      console.log(`   Total tokens: ${usage.totalTokens}`)
+      console.log('   ✅ Token usage tracking works')
+    } else {
+      console.log('   ⚠️  No usage data returned')
+    }
+  } catch (error) {
+    console.error(`   ❌ Error: ${error}`)
+  }
+
+  // Test 4: Concurrent processing performance
+  console.log('\n📝 Test 4: Concurrent Processing Performance')
+  console.log('   Creating 10 items to test concurrency...')
+
+  const concurrentItems: BatchItem[] = Array.from({ length: 10 }, (_, i) => ({
+    id: `concurrent_${i}`,
+    prompt: `What is ${i + 1} * 2? Reply with just the number.`,
+    status: 'pending' as const,
+  }))
+
+  try {
+    const startTime = Date.now()
+    const results = await adapter.submitFlex(concurrentItems, { model: 'gpt-4o-mini' })
+    const duration = Date.now() - startTime
+
+    const successCount = results.filter((r) => r.status === 'completed').length
+    console.log(`   ✅ Processed ${successCount}/10 items in ${duration}ms`)
+    console.log(`   Average: ${Math.round(duration / 10)}ms per item`)
+
+    if (duration < 10000) {
+      console.log('   ✅ Concurrent processing is working (< 10s for 10 items)')
+    } else {
+      console.log('   ⚠️  Slower than expected, may not be fully concurrent')
+    }
+  } catch (error) {
+    console.error(`   ❌ Error: ${error}`)
+  }
+
+  // Test 5: Tier selection
+  console.log('\n📝 Test 5: Execution Tier Selection')
+
+  console.log(`   1 item  → ${getExecutionTier(1)} (expected: immediate)`)
+  console.log(`   3 item  → ${getExecutionTier(3)} (expected: flex, threshold=3)`)
+  console.log(`   10 items → ${getExecutionTier(10)} (expected: flex)`)
+  console.log(`   500 items → ${getExecutionTier(500)} (expected: batch)`)
+
+  console.log('\n✨ E2E Tests Complete!\n')
+
+  resetContext()
+}
+
+main().catch((error) => {
+  console.error('Fatal error:', error)
+  process.exit(1)
+})