# Safe Migration & Testing Plan

## Overview

This document ensures the refactor is done safely with zero breakage. All existing functionality must work identically before and after migration.

---

## Current Working System Components

### 1. AI Functions (Core - Used by ALL paths)

All AI operations go through these functions in `backend/igny8_core/ai/functions/`:

| Function | File | Operation Type | Credits | Called Via |
|----------|------|----------------|---------|------------|
| Auto Cluster | `auto_cluster.py` | `clustering` | Token-based | AIEngine |
| Generate Ideas | `generate_ideas.py` | `idea_generation` | Token-based | AIEngine |
| Generate Content | `generate_content.py` | `content_generation` | Token-based | AIEngine |
| Generate Image Prompts | `generate_image_prompts.py` | `image_prompt_extraction` | Token-based | AIEngine |
| Generate Images | `generate_images.py` | `image_generation` | Per-image | AICore direct |
| Optimize Content | `optimize_content.py` | `content_optimization` | Token-based | OptimizerService |

**Operation Type Mapping** (in `engine.py`):
```python
mapping = {
    'auto_cluster': 'clustering',
    'generate_ideas': 'idea_generation',
    'generate_content': 'content_generation',
    'generate_image_prompts': 'image_prompt_extraction',
    'generate_images': 'image_generation',
    'generate_site_structure': 'site_structure_generation',
}
```

### 1.1 Non-AIEngine Credit Deduction Points

| Service | Operation Type | Direct Credit Deduction |
|---------|---------------|------------------------|
| LinkerService | `internal_linking` | ✅ Uses CreditService directly |
| OptimizerService | `content_optimization` | ✅ Uses CreditService directly |

### 2. Execution Paths (All use same AI functions)

```
┌─────────────────────────────────────────────────────────────────────────┐
│                    ALL EXECUTION PATHS                                   │
└─────────────────────────────────────────────────────────────────────────┘

PATH 1: Manual Module Page Buttons (via run_ai_task Celery task)
├── Planner Page → "Cluster Keywords" button → api/planner.py → run_ai_task.delay('auto_cluster')
├── Planner Page → "Generate Ideas" button → api/planner.py → run_ai_task.delay('generate_ideas')
├── Writer Page → "Generate Content" button → api/writer.py → run_ai_task.delay('generate_content')
├── Writer Page → "Extract Prompts" button → api/writer.py → run_ai_task.delay('generate_image_prompts')
├── Writer Page → "Generate Images" button → api/writer.py → process_image_generation_queue.delay()
└── All go through: Celery task → AIEngine → AI Function → AICore → API → CreditService

PATH 2: Automation Page Manual Run
├── Automation Page → "Run Now" button
├── automation/tasks.py → run_automation_task
├── AutomationService.run_automation()
│   Stage 1: _run_clustering() → AIEngine.execute('auto_cluster')
│   Stage 2: _run_idea_generation() → AIEngine.execute('generate_ideas')
│   Stage 3: _run_task_creation() → Database only (no AI)
│   Stage 4: _run_content_generation() → AIEngine.execute('generate_content')
│   Stage 5: _run_image_prompt_extraction() → AIEngine.execute('generate_image_prompts')
│   Stage 6: _run_image_generation() → process_image_generation_queue.delay()
│   Stage 7: _run_publishing() → WordPress API (no AI)
└── Same credit deduction path (stages 1,2,4,5 via AIEngine, stage 6 via AICore)

PATH 3: Scheduled Automation Run
├── Celery Beat scheduler → check_scheduled_automations task (hourly)
├── automation_tasks.py → run_automation_task.delay()
├── AutomationService.run_automation()
├── Calls same AI functions through AIEngine
└── Same credit deduction path

PATH 4: Direct Service Operations (Bypass AIEngine)
├── Linker Module → LinkerService → CreditService.deduct_credits_for_operation('internal_linking')
├── Optimizer Module → OptimizerService → CreditService.deduct_credits_for_operation('content_optimization')
└── These use direct credit deduction, not via AIEngine
```

### 3. Credit Flow (Current)

```
┌─────────────────────────────────────────────────────────────────────────┐
│                    CURRENT CREDIT FLOW                                   │
└─────────────────────────────────────────────────────────────────────────┘

For TEXT operations (clustering, ideas, content) - VIA AIEngine:
1. API endpoint receives request
2. Celery task run_ai_task triggered
3. AIEngine.execute(function_name, payload)
   a. Step 2.5: CreditService.check_credits() - PRE-CHECK
   b. Step 3: AICore.run_ai_request() - makes OpenAI API call
   c. Response includes token usage (input_tokens, output_tokens)
   d. Step 5.5: CreditService.deduct_credits_for_operation() - DEDUCTION
4. CreditService.calculate_credits_from_tokens() uses CreditCostConfig
5. Creates CreditTransaction (balance history)
6. Creates CreditUsageLog (detailed tracking)
7. Updates Account.credits

For IMAGE operations - VIA process_image_generation_queue:
1. API endpoint receives request
2. Celery task process_image_generation_queue triggered
3. For each image: AICore.generate_image()
   a. Makes Runware/OpenAI API call
   b. Calculates cost via ModelRegistry.calculate_cost()
   c. Returns image URL
4. ⚠️ Credit deduction handled AFTER all images generated
5. Uses CreditCostConfig.image_generation entry (min_credits, tokens_per_credit)
6. Creates CreditUsageLog

For DIRECT SERVICE OPERATIONS (Linker, Optimizer):
1. Service method called (e.g., LinkerService.generate_links())
2. AI call made directly to AICore or external service
3. CreditService.deduct_credits_for_operation() called directly
4. Uses operation-specific CreditCostConfig entry
```

### 3.1 Celery Task Definitions

| Task File | Task Name | Entry Point For |
|-----------|-----------|-----------------|
| `ai/tasks.py` | `run_ai_task` | All manual AI buttons (universal task) |
| `ai/tasks.py` | `process_image_generation_queue` | Image generation queue |
| `automation/tasks.py` | `check_scheduled_automations` | Hourly scheduler check |
| `automation/tasks.py` | `run_automation_task` | Full automation pipeline |
| `automation/tasks.py` | `resume_paused_automation` | Resume after pause |

### 4. Key Files Currently Working

| File | Purpose | Must Keep Working |
|------|---------|-------------------|
| `ai/engine.py` | Orchestrates AI functions | ✅ |
| `ai/ai_core.py` | API calls, key loading | ✅ |
| `ai/model_registry.py` | Model config lookup | ✅ |
| `ai/settings.py` | get_model_config() | ✅ |
| `ai/tasks.py` | Celery tasks | ✅ |
| `ai/functions/*.py` | All AI functions | ✅ |
| `billing/services/credit_service.py` | Credit calculation | ✅ |
| `billing/models.py` | CreditUsageLog, CreditCostConfig | ✅ |
| `automation/services/automation_service.py` | Automation runner | ✅ |
| `modules/system/global_settings_models.py` | API keys, defaults | ✅ |

---

## Potential Issues to Verify Before Migration

### ⚠️ Issue 1: Image Credit Deduction May Be Incomplete

The `process_image_generation_queue` task calculates cost via `ModelRegistry.calculate_cost()` but credit deduction needs verification:
- Does it call `CreditService.deduct_credits_for_operation()`?
- Or does it just log cost without deducting credits?

**Verification Query:**
```bash
docker exec igny8_backend python manage.py shell -c "
from igny8_core.business.billing.models import CreditUsageLog
logs = CreditUsageLog.objects.filter(operation_type='image_generation').order_by('-created_at')[:5]
for log in logs:
    print(f'{log.operation_type}: {log.credits_used} credits, cost=${log.cost_usd}, model={log.model_used}')
if not logs:
    print('⚠️ NO image_generation credit logs found - may not be tracking credits!')
"
```

### ⚠️ Issue 2: GlobalIntegrationSettings.runware_model Points to Non-Existent Model

Current value: `bria:10@1` - but this model doesn't exist in AIModelConfig.
This may cause fallback issues.

### ⚠️ Issue 3: Multiple Credit Calculation Paths

Credits are calculated in different ways depending on the path:
- Via AIEngine → uses `CreditCostConfig` token-based calculation
- Via direct service → uses `CreditCostConfig` with operation-specific rates
- Via image generation → may use different calculation

Need to unify after migration.

---

## Pre-Migration Baseline Capture

### Step 1: Database State Snapshot

Run this BEFORE any changes to establish baseline:

```bash
docker exec igny8_backend python manage.py shell << 'EOF'
import json
from datetime import datetime

print("=" * 80)
print(f"BASELINE SNAPSHOT - {datetime.now().isoformat()}")
print("=" * 80)

# 1. AIModelConfig
print("\n=== AIModelConfig ===")
from igny8_core.business.billing.models import AIModelConfig
for m in AIModelConfig.objects.all().order_by('model_type', 'sort_order'):
    print(json.dumps({
        'model_name': m.model_name,
        'model_type': m.model_type,
        'provider': m.provider,
        'is_active': m.is_active,
        'is_default': m.is_default,
        'cost_per_image': str(m.cost_per_image) if m.cost_per_image else None,
        'input_cost_per_1m': str(m.input_cost_per_1m) if m.input_cost_per_1m else None,
        'output_cost_per_1m': str(m.output_cost_per_1m) if m.output_cost_per_1m else None,
    }))

# 2. CreditCostConfig
print("\n=== CreditCostConfig ===")
from igny8_core.business.billing.models import CreditCostConfig
for c in CreditCostConfig.objects.filter(is_active=True):
    print(json.dumps({
        'operation_type': c.operation_type,
        'tokens_per_credit': c.tokens_per_credit,
        'min_credits': c.min_credits,
        'price_per_credit_usd': str(c.price_per_credit_usd),
    }))

# 3. GlobalIntegrationSettings
print("\n=== GlobalIntegrationSettings ===")
from igny8_core.modules.system.global_settings_models import GlobalIntegrationSettings
g = GlobalIntegrationSettings.get_instance()
print(json.dumps({
    'openai_model': g.openai_model,
    'openai_temperature': g.openai_temperature,
    'openai_max_tokens': g.openai_max_tokens,
    'default_text_provider': g.default_text_provider,
    'dalle_model': g.dalle_model,
    'runware_model': g.runware_model,
    'default_image_service': g.default_image_service,
    'image_style': g.image_style,
    'max_in_article_images': g.max_in_article_images,
    'has_openai_key': bool(g.openai_api_key),
    'has_runware_key': bool(g.runware_api_key),
}))

# 4. Recent Credit Usage (last 20)
print("\n=== Recent CreditUsageLog (last 20) ===")
from igny8_core.business.billing.models import CreditUsageLog
for log in CreditUsageLog.objects.all().order_by('-created_at')[:20]:
    print(json.dumps({
        'operation_type': log.operation_type,
        'credits_used': log.credits_used,
        'model_used': log.model_used,
        'tokens_input': log.tokens_input,
        'tokens_output': log.tokens_output,
        'cost_usd': str(log.cost_usd) if log.cost_usd else None,
    }))

# 5. Account Credits
print("\n=== Account Credits ===")
from igny8_core.auth.models import Account
for acc in Account.objects.all()[:5]:
    print(json.dumps({
        'account_id': acc.id,
        'name': acc.name,
        'credits': acc.credits,
    }))

print("\n" + "=" * 80)
print("BASELINE CAPTURE COMPLETE")
print("=" * 80)
EOF
```

Save this output to: `4th-jan-refactor/baseline-snapshot.json`

---

## Integration Tests (Must Pass Before AND After)

### Test File: `backend/igny8_core/tests/test_ai_system_integration.py`

```python
"""
AI System Integration Tests
===========================
These tests verify the entire AI pipeline works end-to-end.
Run BEFORE migration to establish baseline.
Run AFTER each phase to verify nothing broke.

Usage:
    docker exec igny8_backend python manage.py test igny8_core.tests.test_ai_system_integration
"""
from decimal import Decimal
from django.test import TestCase, TransactionTestCase
from django.db import transaction
from unittest.mock import patch, MagicMock
import json


class AIModelConfigTests(TestCase):
    """Test AIModelConfig loading and access"""
    
    def test_text_models_exist(self):
        """All required text models exist and are configured"""
        from igny8_core.business.billing.models import AIModelConfig
        
        required_text_models = ['gpt-5.1', 'gpt-4o-mini']
        for model_name in required_text_models:
            model = AIModelConfig.objects.filter(model_name=model_name).first()
            self.assertIsNotNone(model, f"Text model {model_name} not found")
            self.assertEqual(model.model_type, 'text')
    
    def test_image_models_exist(self):
        """All required image models exist and are configured"""
        from igny8_core.business.billing.models import AIModelConfig
        
        required_image_models = ['runware:97@1', 'dall-e-3', 'google:4@2']
        for model_name in required_image_models:
            model = AIModelConfig.objects.filter(model_name=model_name).first()
            self.assertIsNotNone(model, f"Image model {model_name} not found")
            self.assertEqual(model.model_type, 'image')
            self.assertIsNotNone(model.cost_per_image, f"{model_name} missing cost_per_image")
    
    def test_default_text_model_exists(self):
        """Exactly one default text model is set"""
        from igny8_core.business.billing.models import AIModelConfig
        
        defaults = AIModelConfig.objects.filter(model_type='text', is_default=True, is_active=True)
        self.assertEqual(defaults.count(), 1, "Should have exactly 1 default text model")
        self.assertEqual(defaults.first().model_name, 'gpt-5.1')
    
    def test_default_image_model_exists(self):
        """Exactly one default image model is set"""
        from igny8_core.business.billing.models import AIModelConfig
        
        defaults = AIModelConfig.objects.filter(model_type='image', is_default=True, is_active=True)
        self.assertEqual(defaults.count(), 1, "Should have exactly 1 default image model")


class ModelRegistryTests(TestCase):
    """Test ModelRegistry functionality"""
    
    def test_get_model_from_db(self):
        """ModelRegistry.get_model() returns model from database"""
        from igny8_core.ai.model_registry import ModelRegistry
        
        model = ModelRegistry.get_model('gpt-5.1')
        self.assertIsNotNone(model)
        # Should be AIModelConfig instance or dict with model_name
        if hasattr(model, 'model_name'):
            self.assertEqual(model.model_name, 'gpt-5.1')
        else:
            self.assertEqual(model.get('model_name'), 'gpt-5.1')
    
    def test_get_image_model(self):
        """ModelRegistry returns image models correctly"""
        from igny8_core.ai.model_registry import ModelRegistry
        
        for model_name in ['runware:97@1', 'dall-e-3', 'google:4@2']:
            model = ModelRegistry.get_model(model_name)
            self.assertIsNotNone(model, f"Model {model_name} not found")
    
    def test_calculate_cost_text(self):
        """Cost calculation for text models works"""
        from igny8_core.ai.model_registry import ModelRegistry
        
        cost = ModelRegistry.calculate_cost('gpt-5.1', input_tokens=1000, output_tokens=500)
        self.assertIsInstance(cost, (int, float, Decimal))
        self.assertGreater(cost, 0)
    
    def test_calculate_cost_image(self):
        """Cost calculation for image models works"""
        from igny8_core.ai.model_registry import ModelRegistry
        
        cost = ModelRegistry.calculate_cost('dall-e-3', num_images=1)
        self.assertIsInstance(cost, (int, float, Decimal))
        self.assertGreater(cost, 0)


class APIKeyLoadingTests(TestCase):
    """Test API key loading from GlobalIntegrationSettings"""
    
    def test_openai_key_loads(self):
        """OpenAI API key loads from GlobalIntegrationSettings"""
        from igny8_core.ai.ai_core import AICore
        
        ai_core = AICore()
        key = ai_core.get_api_key('openai')
        self.assertIsNotNone(key, "OpenAI API key not configured")
        self.assertTrue(len(key) > 10, "OpenAI API key too short")
    
    def test_runware_key_loads(self):
        """Runware API key loads from GlobalIntegrationSettings"""
        from igny8_core.ai.ai_core import AICore
        
        ai_core = AICore()
        key = ai_core.get_api_key('runware')
        self.assertIsNotNone(key, "Runware API key not configured")


class CreditServiceTests(TestCase):
    """Test credit calculation and deduction"""
    
    def test_calculate_credits_text_operation(self):
        """Credit calculation for text operations works"""
        from igny8_core.business.billing.services.credit_service import CreditService
        
        # Test content generation
        credits = CreditService.calculate_credits_from_tokens(
            'content_generation',
            tokens_input=1000,
            tokens_output=2000
        )
        self.assertIsInstance(credits, int)
        self.assertGreater(credits, 0)
    
    def test_calculate_credits_image_operation(self):
        """Credit calculation for image operations works"""
        from igny8_core.business.billing.services.credit_service import CreditService
        
        # Image generation uses min_credits from CreditCostConfig
        credits = CreditService.calculate_credits_from_tokens(
            'image_generation',
            tokens_input=0,
            tokens_output=0
        )
        self.assertIsInstance(credits, int)
        self.assertGreaterEqual(credits, 1)
    
    def test_credit_cost_config_exists(self):
        """All required CreditCostConfig entries exist"""
        from igny8_core.business.billing.models import CreditCostConfig
        
        required_ops = ['clustering', 'idea_generation', 'content_generation', 
                       'image_generation', 'image_prompt_extraction']
        for op in required_ops:
            config = CreditCostConfig.objects.filter(operation_type=op, is_active=True).first()
            self.assertIsNotNone(config, f"CreditCostConfig for {op} not found")


class AISettingsTests(TestCase):
    """Test get_model_config() function"""
    
    def test_get_model_config_returns_config(self):
        """get_model_config() returns valid configuration"""
        from igny8_core.ai.settings import get_model_config
        from igny8_core.auth.models import Account
        
        account = Account.objects.first()
        if account:
            config = get_model_config('content_generation', account)
            
            self.assertIn('model', config)
            self.assertIn('max_tokens', config)
            self.assertIn('temperature', config)
            self.assertIsNotNone(config['model'])


class AIFunctionValidationTests(TestCase):
    """Test AI function classes load correctly"""
    
    def test_auto_cluster_function_loads(self):
        """AutoClusterFunction loads and validates"""
        from igny8_core.ai.functions.auto_cluster import AutoClusterFunction
        
        fn = AutoClusterFunction()
        self.assertEqual(fn.get_name(), 'auto_cluster')
        self.assertIsNotNone(fn.get_metadata())
    
    def test_generate_ideas_function_loads(self):
        """GenerateIdeasFunction loads and validates"""
        from igny8_core.ai.functions.generate_ideas import GenerateIdeasFunction
        
        fn = GenerateIdeasFunction()
        self.assertEqual(fn.get_name(), 'generate_ideas')
    
    def test_generate_content_function_loads(self):
        """GenerateContentFunction loads and validates"""
        from igny8_core.ai.functions.generate_content import GenerateContentFunction
        
        fn = GenerateContentFunction()
        self.assertEqual(fn.get_name(), 'generate_content')
    
    def test_generate_images_function_loads(self):
        """GenerateImagesFunction loads and validates"""
        from igny8_core.ai.functions.generate_images import GenerateImagesFunction
        
        fn = GenerateImagesFunction()
        self.assertEqual(fn.get_name(), 'generate_images')


class GlobalIntegrationSettingsTests(TestCase):
    """Test GlobalIntegrationSettings singleton"""
    
    def test_singleton_loads(self):
        """GlobalIntegrationSettings singleton loads"""
        from igny8_core.modules.system.global_settings_models import GlobalIntegrationSettings
        
        settings = GlobalIntegrationSettings.get_instance()
        self.assertIsNotNone(settings)
        self.assertEqual(settings.pk, 1)
    
    def test_image_settings_exist(self):
        """Image settings are configured"""
        from igny8_core.modules.system.global_settings_models import GlobalIntegrationSettings
        
        settings = GlobalIntegrationSettings.get_instance()
        self.assertIsNotNone(settings.image_style)
        self.assertIsNotNone(settings.max_in_article_images)
        self.assertGreater(settings.max_in_article_images, 0)


class AutomationServiceTests(TestCase):
    """Test AutomationService uses same AI functions"""
    
    def test_automation_service_imports(self):
        """AutomationService can import AI functions"""
        from igny8_core.business.automation.services.automation_service import AutomationService
        
        # Just verify the import works and class exists
        self.assertIsNotNone(AutomationService)
```

### Running Tests

```bash
# Run all integration tests
docker exec igny8_backend python manage.py test igny8_core.tests.test_ai_system_integration -v 2

# Run specific test class
docker exec igny8_backend python manage.py test igny8_core.tests.test_ai_system_integration.AIModelConfigTests -v 2

# Run with coverage
docker exec igny8_backend coverage run manage.py test igny8_core.tests.test_ai_system_integration
docker exec igny8_backend coverage report
```

---

## Phased Migration with Verification

### Phase 1: Add New Models (NO BREAKING CHANGES)

**Changes:**
- Add `IntegrationProvider` model
- Add new fields to `AIModelConfig`: `credits_per_image`, `tokens_per_credit`, `quality_tier`
- Create migrations
- Populate new data

**Verification:**
```bash
# 1. Run migrations
docker exec igny8_backend python manage.py makemigrations
docker exec igny8_backend python manage.py migrate

# 2. Run ALL tests - must pass
docker exec igny8_backend python manage.py test igny8_core.tests.test_ai_system_integration -v 2

# 3. Manual verification - existing features still work
# - Go to Writer page, generate content (check credits deducted)
# - Go to Writer page, generate images (check credits deducted)
# - Run automation manually (check it completes)
```

**Rollback:** Delete new tables/fields (no existing data affected)

---

### Phase 2: Add Parallel Code Paths (OLD KEEPS WORKING)

**Changes:**
- Add new methods to `ModelRegistry` (keep old ones)
- Add new methods to `CreditService` (keep old ones)
- Add comparison logging

**Pattern:**
```python
# model_registry.py
@classmethod
def get_default_model_NEW(cls, model_type: str):
    """NEW: Get default model from AIModelConfig.is_default"""
    from igny8_core.business.billing.models import AIModelConfig
    return AIModelConfig.objects.filter(
        model_type=model_type, 
        is_default=True, 
        is_active=True
    ).first()

@classmethod  
def get_model(cls, model_id: str):
    """EXISTING: Keep working exactly as before"""
    # ... existing code unchanged ...
```

```python
# credit_service.py
@staticmethod
def calculate_credits_for_image_NEW(model_name: str, num_images: int) -> int:
    """NEW: Calculate from AIModelConfig.credits_per_image"""
    from igny8_core.business.billing.models import AIModelConfig
    try:
        model = AIModelConfig.objects.get(model_name=model_name, is_active=True)
        if model.credits_per_image:
            return model.credits_per_image * num_images
    except AIModelConfig.DoesNotExist:
        pass
    return None  # Signal to use old method

@staticmethod
def calculate_credits_from_tokens(operation_type, tokens_input, tokens_output):
    """EXISTING: Keep working, add comparison logging"""
    # Calculate old way
    old_result = cls._calculate_old(operation_type, tokens_input, tokens_output)
    
    # Calculate new way (for comparison only)
    new_result = cls._calculate_new(operation_type, tokens_input, tokens_output)
    
    # Log if different
    if old_result != new_result:
        logger.warning(f"[MIGRATION] Credit calc mismatch: {operation_type} old={old_result} new={new_result}")
    
    # Return OLD (safe) for now
    return old_result
```

**Verification:**
```bash
# 1. Run tests
docker exec igny8_backend python manage.py test igny8_core.tests.test_ai_system_integration -v 2

# 2. Check logs for any mismatch warnings
docker logs igny8_backend 2>&1 | grep "MIGRATION"

# 3. Manual test all paths
```

**Rollback:** Remove new methods only

---

### Phase 3: Add IntegrationProvider for API Keys (OLD FALLBACK)

**Changes:**
- `AICore._load_account_settings()` tries `IntegrationProvider` first, falls back to `GlobalIntegrationSettings`

**Pattern:**
```python
# ai_core.py
def _load_account_settings(self):
    """Load API keys - try new IntegrationProvider, fallback to old"""
    
    # Try NEW way first
    try:
        from igny8_core.business.billing.models import IntegrationProvider
        openai_provider = IntegrationProvider.objects.filter(
            provider_id='openai', is_active=True
        ).first()
        if openai_provider and openai_provider.api_key:
            self._openai_api_key = openai_provider.api_key
            logger.info("[MIGRATION] Loaded OpenAI key from IntegrationProvider")
        else:
            raise Exception("Fallback to old method")
    except Exception:
        # FALLBACK to old GlobalIntegrationSettings
        from igny8_core.modules.system.global_settings_models import GlobalIntegrationSettings
        global_settings = GlobalIntegrationSettings.get_instance()
        self._openai_api_key = global_settings.openai_api_key
        logger.info("[MIGRATION] Loaded OpenAI key from GlobalIntegrationSettings (fallback)")
    
    # Same pattern for runware...
```

**Verification:**
```bash
# 1. Run tests
docker exec igny8_backend python manage.py test igny8_core.tests.test_ai_system_integration -v 2

# 2. Verify API calls work
# - Generate content (uses OpenAI)
# - Generate images with Runware model
# - Generate images with DALL-E model

# 3. Check logs
docker logs igny8_backend 2>&1 | grep "MIGRATION"
```

**Rollback:** Revert `_load_account_settings()` to old version

---

### Phase 4: Switch to New Credit Calculation

**Changes:**
- Use `AIModelConfig.credits_per_image` for image operations
- Use `AIModelConfig.tokens_per_credit` for text operations
- Keep `CreditCostConfig` as fallback

**Verification:**
```bash
# 1. Before switching, compare calculations for 1 week
# Log both old and new results, verify they match

# 2. Run tests
docker exec igny8_backend python manage.py test igny8_core.tests.test_ai_system_integration -v 2

# 3. Verify credit logs show correct values
docker exec igny8_backend python manage.py shell -c "
from igny8_core.business.billing.models import CreditUsageLog
for log in CreditUsageLog.objects.order_by('-created_at')[:10]:
    print(f'{log.operation_type}: {log.credits_used} credits, model={log.model_used}')
"
```

---

### Phase 5: Create API Endpoint

**Changes:**
- Add `/api/v1/system/ai-models/` endpoint
- Returns models from database

**Verification:**
```bash
# 1. Test endpoint
curl -H "Authorization: Bearer $TOKEN" http://localhost:8000/api/v1/system/ai-models/

# 2. Verify response structure
# Should have text_models, image_models, image_settings
```

---

### Phase 6: Update Frontend

**Changes:**
- Remove hardcoded model choices from `Settings.tsx`
- Load from API

**Verification:**
```bash
# 1. Frontend tests
cd frontend && npm run test

# 2. Manual verification
# - Open Sites > Settings > Image Settings
# - Verify dropdown shows models from API
# - Change model, save, verify it persists
```

---

### Phase 7: Cleanup (After 1 Week Stable)

**Remove from GlobalIntegrationSettings:**
```python
# API key fields (moved to IntegrationProvider)
- openai_api_key
- anthropic_api_key  # unused
- bria_api_key       # unused
- runware_api_key

# Model selection fields (now from AIModelConfig.is_default)
- openai_model
- dalle_model
- runware_model
- default_text_provider
- default_image_service

# Hardcoded CHOICES
- OPENAI_MODEL_CHOICES
- DALLE_MODEL_CHOICES
- RUNWARE_MODEL_CHOICES
- TEXT_PROVIDER_CHOICES
- IMAGE_SERVICE_CHOICES
```

**Remove from constants.py:**
```python
- MODEL_RATES = {...}
- IMAGE_MODEL_RATES = {...}
```

**Remove from model_registry.py:**
```python
- Fallback to constants.py (all lookups must use DB)
- Hardcoded default model names
```

**Remove from ai_core.py:**
```python
- IMAGE_MODEL_RATES import and usage
- Direct GlobalIntegrationSettings key access (use IntegrationProvider)
```

**Remove CreditCostConfig entries (optional - use AIModelConfig):**
```python
- image_generation entry (use AIModelConfig.credits_per_image instead)
```

**Remove from frontend Settings.tsx:**
```python
- QUALITY_TO_CONFIG hardcoded mapping
- RUNWARE_MODEL_CHOICES hardcoded array
- DALLE_MODEL_CHOICES hardcoded array
- MODEL_LANDSCAPE_SIZES hardcoded mapping
- Any hardcoded model names
```

**Verification:**
```bash
# 1. Run full test suite
docker exec igny8_backend python manage.py test

# 2. Run integration tests
docker exec igny8_backend python manage.py test igny8_core.tests.test_ai_system_integration -v 2

# 3. Full manual test of all paths
```

---

## Complete Test Checklist

### Manual Test Checklist (Run After Each Phase)

#### Path 1: Manual Module Page Actions

| Test | Page | Button/Action | Expected Result | Verify Credits | Pass? |
|------|------|---------------|-----------------|----------------|-------|
| Cluster Keywords | Planner | "Cluster" button | Creates clusters | Check CreditUsageLog for `clustering` | ☐ |
| Generate Ideas | Planner | "Generate Ideas" button | Creates ideas | Check CreditUsageLog for `idea_generation` | ☐ |
| Generate Content | Writer | "Generate" button | Creates content | Check CreditUsageLog for `content_generation` | ☐ |
| Generate Product Content | Writer | "Generate Product" button | Creates product content | Check CreditUsageLog for `content_generation` | ☐ |
| Generate Service Content | Writer | "Generate Service" button | Creates service content | Check CreditUsageLog for `content_generation` | ☐ |
| Extract Image Prompts | Writer | "Extract Prompts" button | Creates prompts | Check CreditUsageLog for `image_prompt_extraction` | ☐ |
| Generate Images (Basic) | Writer | Image gen w/ Runware basic | Creates images | Check CreditUsageLog: 1 credit/image | ☐ |
| Generate Images (Quality) | Writer | Image gen w/ DALL-E | Creates images | Check CreditUsageLog: 5 credits/image | ☐ |
| Generate Images (Premium) | Writer | Image gen w/ Google | Creates images | Check CreditUsageLog: 15 credits/image | ☐ |
| Internal Linking | Linker | "Generate Links" button | Creates links | Check CreditUsageLog for `internal_linking` | ☐ |
| Optimize Content | Optimizer | "Optimize" button | Optimizes content | Check CreditUsageLog for `content_optimization` | ☐ |

#### Path 2: Automation Manual Run

| Test | Page | Action | Expected Result | Verify Credits | Pass? |
|------|------|--------|-----------------|----------------|-------|
| Run Full Automation | Automation | "Run Now" button | All stages complete | Credits for each stage | ☐ |
| Run Stage 1 Only | Automation | Run clustering stage | Clusters created | Check `clustering` credits | ☐ |
| Run Stage 2 Only | Automation | Run idea stage | Ideas created | Check `idea_generation` credits | ☐ |
| Run Stage 4 Only | Automation | Run content stage | Content created | Check `content_generation` credits | ☐ |
| Run Stage 5 Only | Automation | Run prompts stage | Prompts created | Check `image_prompt_extraction` credits | ☐ |
| Run Stage 6 Only | Automation | Run image stage | Images created | Check `image_generation` credits | ☐ |
| Pause/Resume | Automation | Pause then resume | Continues correctly | No double-charging | ☐ |

#### Path 3: Scheduled Automation

| Test | Setup | Expected | Verify | Pass? |
|------|-------|----------|--------|-------|
| Schedule Triggers | Set schedule for automation | Runs on schedule | Check logs at scheduled time | ☐ |
| Credits Deducted | After scheduled run | Credits reduced | Check account balance | ☐ |
| Multiple Automations | Multiple scheduled | All run | Each deducts credits | ☐ |

#### Credit Verification Queries

```bash
# Check recent credit usage logs
docker exec igny8_backend python manage.py shell -c "
from igny8_core.business.billing.models import CreditUsageLog
for log in CreditUsageLog.objects.order_by('-created_at')[:20]:
    print(f'{log.created_at}: {log.operation_type} | {log.credits_used} credits | model={log.model_used} | tokens={log.tokens_input}+{log.tokens_output}')
"

# Check account balance
docker exec igny8_backend python manage.py shell -c "
from igny8_core.auth.models import Account
for acc in Account.objects.all()[:5]:
    print(f'{acc.name}: {acc.credits} credits')
"

# Check credit transactions
docker exec igny8_backend python manage.py shell -c "
from igny8_core.business.billing.models import CreditTransaction
for txn in CreditTransaction.objects.order_by('-created_at')[:10]:
    print(f'{txn.created_at}: {txn.transaction_type} | {txn.credits_amount} | balance={txn.balance_after}')
"
```

#### Frontend Verification

| Test | Page | Action | Expected | Pass? |
|------|------|--------|----------|-------|
| Settings Load | Sites > Settings | Open page | Image settings dropdown populated | ☐ |
| Model Selection | Sites > Settings | Change image model | Shows Basic/Quality/Premium | ☐ |
| Model Persists | Sites > Settings | Save and reload | Selected model persists | ☐ |
| Credits Display | Any page | Header/sidebar | Shows correct credit balance | ☐ |
| Usage Analytics | Analytics | View usage | Shows correct breakdown | ☐ |

---

## Rollback Procedures

### If Phase 1 Fails
```bash
# Remove new migrations
docker exec igny8_backend python manage.py migrate billing <previous_migration>
docker exec igny8_backend python manage.py migrate system <previous_migration>
```

### If Phase 2-6 Fails
```bash
# Revert code changes via git
git checkout -- backend/igny8_core/ai/
git checkout -- backend/igny8_core/business/billing/

# Restart containers
docker-compose restart backend
```

### If Phase 7 Fails (Cleanup)
```bash
# This is why we wait 1 week before cleanup
# Restore from backup or revert git
```

---

## Success Criteria

Migration is COMPLETE when:

1. ✅ All integration tests pass
2. ✅ All manual tests pass
3. ✅ No errors in logs for 1 week
4. ✅ Credit calculations match expected values
5. ✅ All execution paths work (manual, automation, scheduled)
6. ✅ Frontend loads models from API
7. ✅ No legacy code remains:
   - No `MODEL_RATES` / `IMAGE_MODEL_RATES` in constants.py
   - No API keys in `GlobalIntegrationSettings`
   - No hardcoded models in frontend
   - No fallback to constants in `ModelRegistry`