# IGNY8 Phase 2: Content Optimizer (02F)
## Cluster-Aligned Content Optimization Engine

**Document Version:** 1.0
**Date:** 2026-03-23
**Phase:** IGNY8 Phase 2 — Feature Expansion
**Status:** Build Ready
**Source of Truth:** Codebase at `/data/app/igny8/`
**Audience:** Claude Code, Backend Developers, Architects

---

## 1. CURRENT STATE

### Optimization App Today
The `optimization` Django app exists in `INSTALLED_APPS` but is **inactive** (behind feature flag). The following exist:

- **`OptimizationTask` model** — exists with minimal fields (basic task tracking only)
- **`optimize_content` AI function** — registered in `igny8_core/ai/registry.py` as one of the 7 registered functions, but only does basic content rewriting without cluster awareness, keyword coverage analysis, or scoring
- **`optimization` app label** — app exists at `igny8_core/modules/optimization/`

### What Does Not Exist
- No cluster-alignment during optimization
- No keyword coverage analysis against cluster keyword sets
- No heading restructure logic
- No intent-based content rewrite
- No schema gap detection
- No before/after scoring system (0-100)
- No batch optimization
- No integration with SAG data (01A) or taxonomy terms (02B)

### Foundation Available
- `Clusters` model (app_label=`planner`, db_table=`igny8_clusters`) with cluster keywords
- `Keywords` model (app_label=`planner`, db_table=`igny8_keywords`) linked to clusters
- `Content.schema_markup` JSONField — used by 02G for JSON-LD
- `Content.content_type` and `Content.content_structure` — routing context
- `Content.structured_data` JSONField (added by 02A)
- `ContentTaxonomy` cluster mapping (added by 02B) with `mapping_confidence`
- `GSCMetricsCache` (added by 02C) — position data identifies pages needing optimization
- `SchemaValidationService` (added by 02G) — schema gap detection reuse
- `BaseAIFunction` with `validate()`, `prepare()`, `build_prompt()`, `parse_response()`, `save_output()`

---

## 2. WHAT TO BUILD

### Overview
Extend the existing `OptimizationTask` model and `optimize_content` AI function into a full cluster-aligned optimization engine. The system analyzes content against its cluster's keyword set, scores quality on a 0-100 scale, and produces optimized content with tracked before/after metrics.

### 2.1 Cluster Matching (Auto-Assign Optimization Context)

When content has no cluster assignment, the optimizer auto-detects the best-fit cluster:

**Scoring Algorithm:**
- Keyword overlap (40%): count of cluster keywords found in content title + headings + body
- Semantic similarity (40%): AI-scored relevance between content topic and cluster theme
- Title match (20%): similarity between content title and cluster name/keywords

**Thresholds:**
- Confidence ≥ 0.6 → auto-assign cluster
- Confidence < 0.6 → flag for manual review, suggest top 3 candidates

This reuses the same scoring pattern as `ClusterMappingService` from 02B.

### 2.2 Keyword Coverage Analysis

For content with an assigned cluster:

1. Load all `Keywords` records belonging to that cluster
2. Scan `content_html` for each keyword: exact match, partial match (stemmed), semantic presence
3. Report per keyword: `{keyword, target_density, current_density, status: present|missing|low_density}`
4. Coverage targets:
   - Hub content (`cluster_hub`): 70%+ of cluster keywords covered
   - Supporting articles: 40%+ of cluster keywords covered
   - Product/service pages: 30%+ (focused on commercial keywords)

### 2.3 Heading Restructure

Analyze H1/H2/H3 hierarchy for SEO best practices:

| Check | Rule | Fix |
|-------|------|-----|
| Single H1 | Content must have exactly one H1 | Merge or demote extra H1s |
| H2 keyword coverage | H2s should contain target keywords from cluster | AI rewrites H2s with keyword incorporation |
| Logical hierarchy | No skipped levels (H1 → H3 without H2) | Insert missing levels |
| H2 count | Minimum 3 H2s for content >1000 words | AI suggests additional H2 sections |
| Missing keyword themes | Cluster keywords not represented in any heading | AI suggests new H2/H3 sections for missing themes |

### 2.4 Content Rewrite (Intent-Aligned)

**Intent Classification:**
- **Informational**: expand explanations, add examples, increase depth, add definitions
- **Commercial**: add comparison tables, pros/cons, feature highlights, trust signals
- **Transactional**: strengthen CTAs, add urgency, streamline conversion path, social proof

**Content Adjustments:**
- Expand thin content (<500 words) to minimum viable length for the content structure
- Compress bloated content (detect and remove redundancy)
- Add missing sections identified by keyword coverage analysis
- Maintain existing tone and style while improving SEO alignment

### 2.5 Schema Gap Detection

Leverages `SchemaValidationService` from 02G:

1. Check existing `Content.schema_markup` against expected schemas for the content type
2. Expected schema by type: Article (post), Product (product), Service (service_page), FAQPage (if FAQ detected), BreadcrumbList (all), HowTo (if steps detected)
3. Identify missing required fields per schema type
4. Generate corrected/complete schema JSON-LD
5. Schema-only optimization mode available (no content rewrite, just schema fix)

### 2.6 Before/After Scoring

**Content Quality Score (0-100):**

| Factor | Weight | Score Criteria |
|--------|--------|---------------|
| Keyword Coverage | 30% | % of cluster keywords present vs target |
| Heading Structure | 20% | Single H1, keyword H2s, logical hierarchy, no skipped levels |
| Content Depth | 20% | Word count vs structure minimum, section completeness, detail level |
| Readability | 15% | Sentence length, paragraph length, Flesch-Kincaid approximation |
| Schema Completeness | 15% | Required schema fields present, validation passes |

Every optimization records `score_before` and `score_after`. Dashboard aggregates show average improvement across all optimizations.

### 2.7 Batch Optimization

- Select content by: cluster ID, score threshold (e.g., all content scoring < 50), content type, date range
- Queue as Celery tasks with priority ordering (lowest scores first)
- Concurrency: max 3 concurrent optimization tasks per account
- Progress tracking via OptimizationTask status field
- Cancel capability: change status to `rejected` to stop processing

---

## 3. DATA MODELS & APIS

### 3.1 Modified Model — OptimizationTask (optimization app)

Extend the existing `OptimizationTask` model with 16 new fields:

```python
# Add to existing OptimizationTask model:

content = models.ForeignKey(
    'writer.Content',
    on_delete=models.CASCADE,
    related_name='optimization_tasks'
)
primary_cluster = models.ForeignKey(
    'planner.Clusters',
    on_delete=models.SET_NULL,
    null=True,
    blank=True,
    related_name='optimization_tasks'
)
secondary_clusters = models.JSONField(
    default=list,
    blank=True,
    help_text='List of Clusters IDs for secondary relevance'
)
keyword_targets = models.JSONField(
    default=list,
    blank=True,
    help_text='[{keyword, target_density, current_density, status}]'
)
optimization_type = models.CharField(
    max_length=20,
    choices=[
        ('full_rewrite', 'Full Rewrite'),
        ('heading_only', 'Heading Only'),
        ('schema_only', 'Schema Only'),
        ('keyword_coverage', 'Keyword Coverage'),
        ('batch', 'Batch'),
    ],
    default='full_rewrite'
)
intent_classification = models.CharField(
    max_length=15,
    choices=[
        ('informational', 'Informational'),
        ('commercial', 'Commercial'),
        ('transactional', 'Transactional'),
    ],
    blank=True,
    default=''
)
score_before = models.FloatField(null=True, blank=True)
score_after = models.FloatField(null=True, blank=True)
content_before = models.TextField(
    blank=True,
    default='',
    help_text='Snapshot of original content_html'
)
content_after = models.TextField(
    blank=True,
    default='',
    help_text='Optimized HTML (null until optimization completes)'
)
metadata_before = models.JSONField(
    default=dict,
    blank=True,
    help_text='{meta_title, meta_description, headings[]}'
)
metadata_after = models.JSONField(
    default=dict,
    blank=True
)
schema_before = models.JSONField(default=dict, blank=True)
schema_after = models.JSONField(default=dict, blank=True)
structure_changes = models.JSONField(
    default=list,
    blank=True,
    help_text='[{change_type, description, before, after}]'
)
confidence_score = models.FloatField(
    null=True,
    blank=True,
    help_text='AI confidence in the quality of changes (0-1)'
)
applied = models.BooleanField(default=False)
applied_at = models.DateTimeField(null=True, blank=True)
```

**Update STATUS choices on OptimizationTask:**
```python
STATUS_CHOICES = [
    ('pending', 'Pending'),
    ('analyzing', 'Analyzing'),
    ('optimizing', 'Optimizing'),
    ('review', 'Ready for Review'),
    ('applied', 'Applied'),
    ('rejected', 'Rejected'),
]
```

**PK:** BigAutoField (integer) — existing model
**Table:** existing `igny8_optimization_tasks` table (no rename needed)

### 3.2 Migration

Single migration in the optimization app (or igny8_core migrations):

```
igny8_core/migrations/XXXX_extend_optimization_task.py
```

**Operations:**
1. `AddField('OptimizationTask', 'content', ...)` — FK to Content
2. `AddField('OptimizationTask', 'primary_cluster', ...)` — FK to Clusters
3. `AddField('OptimizationTask', 'secondary_clusters', ...)` — JSONField
4. `AddField('OptimizationTask', 'keyword_targets', ...)` — JSONField
5. `AddField('OptimizationTask', 'optimization_type', ...)` — CharField
6. `AddField('OptimizationTask', 'intent_classification', ...)` — CharField
7. `AddField('OptimizationTask', 'score_before', ...)` — FloatField
8. `AddField('OptimizationTask', 'score_after', ...)` — FloatField
9. `AddField('OptimizationTask', 'content_before', ...)` — TextField
10. `AddField('OptimizationTask', 'content_after', ...)` — TextField
11. `AddField('OptimizationTask', 'metadata_before', ...)` — JSONField
12. `AddField('OptimizationTask', 'metadata_after', ...)` — JSONField
13. `AddField('OptimizationTask', 'schema_before', ...)` — JSONField
14. `AddField('OptimizationTask', 'schema_after', ...)` — JSONField
15. `AddField('OptimizationTask', 'structure_changes', ...)` — JSONField
16. `AddField('OptimizationTask', 'confidence_score', ...)` — FloatField
17. `AddField('OptimizationTask', 'applied', ...)` — BooleanField
18. `AddField('OptimizationTask', 'applied_at', ...)` — DateTimeField

### 3.3 API Endpoints

All endpoints under `/api/v1/optimizer/`:

| Method | Path | Description |
|--------|------|-------------|
| POST | `/api/v1/optimizer/analyze/` | Analyze single content piece. Body: `{content_id}`. Returns scores + keyword coverage + heading analysis + recommendations. Does NOT rewrite. |
| POST | `/api/v1/optimizer/optimize/` | Run full optimization. Body: `{content_id, optimization_type}`. Creates OptimizationTask, runs analysis + rewrite, returns preview. |
| POST | `/api/v1/optimizer/preview/` | Preview changes without creating task. Body: `{content_id}`. Returns diff-style output. |
| POST | `/api/v1/optimizer/apply/{id}/` | Apply optimized version. Copies `content_after` → `Content.content_html`, updates metadata, sets `applied=True`. |
| POST | `/api/v1/optimizer/reject/{id}/` | Reject optimization. Sets status=`rejected`, keeps original content. |
| POST | `/api/v1/optimizer/batch/` | Queue batch optimization. Body: `{site_id, cluster_id?, score_threshold?, content_type?, content_ids?}`. Returns batch task ID. |
| GET | `/api/v1/optimizer/tasks/?site_id=X` | List OptimizationTask records with filters (status, optimization_type, cluster_id, date range). |
| GET | `/api/v1/optimizer/tasks/{id}/` | Single optimization detail with full before/after data. |
| GET | `/api/v1/optimizer/tasks/{id}/diff/` | HTML diff view — visual comparison of content_before vs content_after. |
| GET | `/api/v1/optimizer/cluster-suggestions/?content_id=X` | Suggest best-fit cluster for unassigned content. Returns top 3 candidates with confidence scores. |
| POST | `/api/v1/optimizer/assign-cluster/` | Assign cluster to content. Body: `{content_id, cluster_id}`. Updates Content record. |
| GET | `/api/v1/optimizer/dashboard/?site_id=X` | Optimization stats: avg score improvement, count by status, top improved, lowest scoring content. |

**Permissions:** All endpoints use `SiteSectorModelViewSet` permission patterns.

### 3.4 AI Function — Enhanced optimize_content

Extend the existing registered `optimize_content` AI function:

**Registry key:** `optimize_content` (already registered — enhance, not replace)
**Location:** `igny8_core/ai/functions/optimize_content.py` (existing file)

```python
class OptimizeContentFunction(BaseAIFunction):
    """
    Enhanced cluster-aligned content optimization.
    Extends existing optimize_content with keyword coverage,
    heading restructure, intent classification, and scoring.
    """
    function_name = 'optimize_content'

    def validate(self, content_id, optimization_type='full_rewrite', **kwargs):
        # Verify content exists, has content_html
        # Verify optimization_type is valid
        pass

    def prepare(self, content_id, optimization_type='full_rewrite', **kwargs):
        # Load Content record
        # Determine cluster (from Content or auto-match)
        # Load cluster Keywords
        # Analyze current keyword coverage
        # Parse heading structure
        # Classify intent
        # Calculate score_before
        # Snapshot content_before, metadata_before, schema_before
        pass

    def build_prompt(self):
        # Build type-specific optimization prompt:
        # - Include current content_html
        # - Include cluster keywords with coverage status
        # - Include heading analysis results
        # - Include intent classification
        # - Include optimization_type instructions:
        #   full_rewrite: all optimizations
        #   heading_only: heading restructure only
        #   schema_only: schema fix only (no content change)
        #   keyword_coverage: add missing keyword sections only
        pass

    def parse_response(self, response):
        # Parse optimized HTML
        # Parse updated metadata (meta_title, meta_description)
        # Parse structure_changes list
        # Parse confidence_score
        pass

    def save_output(self, parsed):
        # Create OptimizationTask with all before/after data
        # Calculate score_after
        # Set status='review'
        pass
```

### 3.5 Content Scoring Service

**Location:** `igny8_core/business/content_scoring.py`

```python
class ContentScoringService:
    """
    Calculates Content Quality Score (0-100) using 5 weighted factors.
    Used by optimizer for before/after and by dashboard for overview.
    """

    WEIGHTS = {
        'keyword_coverage': 0.30,
        'heading_structure': 0.20,
        'content_depth': 0.20,
        'readability': 0.15,
        'schema_completeness': 0.15,
    }

    def score(self, content_id, cluster_id=None):
        """
        Calculate composite score for a content record.
        Returns: {total: float, breakdown: {factor: score}}
        """
        pass

    def _score_keyword_coverage(self, content, cluster):
        """0-100: % of cluster keywords found in content."""
        pass

    def _score_heading_structure(self, content_html):
        """0-100: single H1, keyword H2s, no skipped levels, H2 count."""
        pass

    def _score_content_depth(self, content_html, content_structure):
        """0-100: word count vs minimum for structure type, section completeness."""
        pass

    def _score_readability(self, content_html):
        """0-100: avg sentence length, paragraph length, Flesch-Kincaid approx."""
        pass

    def _score_schema_completeness(self, content):
        """0-100: required schema fields present, from SchemaValidationService (02G)."""
        pass
```

### 3.6 Keyword Coverage Analyzer

**Location:** `igny8_core/business/keyword_coverage.py`

```python
class KeywordCoverageAnalyzer:
    """
    Analyzes content against cluster keyword set.
    Returns per-keyword presence and overall coverage percentage.
    """

    def analyze(self, content_id, cluster_id):
        """
        Returns {
            total_keywords: int,
            covered: int,
            missing: int,
            coverage_pct: float,
            keywords: [{keyword, target_density, current_density, status}]
        }
        """
        pass

    def _extract_text(self, content_html):
        """Strip HTML, return plain text for analysis."""
        pass

    def _check_keyword(self, keyword, text):
        """Check for exact, partial (stemmed), and semantic presence."""
        pass
```

### 3.7 Celery Tasks

**Location:** `igny8_core/tasks/optimization_tasks.py`

```python
@shared_task(name='run_optimization')
def run_optimization(optimization_task_id):
    """Process a single OptimizationTask. Called by API endpoints."""
    pass

@shared_task(name='run_batch_optimization')
def run_batch_optimization(site_id, cluster_id=None, score_threshold=None,
                           content_type=None, content_ids=None, batch_size=10):
    """
    Process batch of content for optimization.
    Selects content matching filters, creates OptimizationTask per item,
    processes sequentially with max 3 concurrent per account.
    """
    pass

@shared_task(name='identify_optimization_candidates')
def identify_optimization_candidates(site_id, threshold=50):
    """
    Weekly scan: find content with quality score below threshold.
    Creates report, does NOT auto-optimize.
    """
    pass
```

**Beat Schedule Addition:**

| Task | Schedule | Notes |
|------|----------|-------|
| `identify_optimization_candidates` | Weekly (Monday 4:00 AM) | Scans all sites, identifies low-scoring content |

---

## 4. IMPLEMENTATION STEPS

### Step 1: Migration
1. Add 16 new fields to `OptimizationTask` model
2. Update STATUS_CHOICES on OptimizationTask
3. Run migration

### Step 2: Services
1. Implement `ContentScoringService` in `igny8_core/business/content_scoring.py`
2. Implement `KeywordCoverageAnalyzer` in `igny8_core/business/keyword_coverage.py`

### Step 3: AI Function Enhancement
1. Extend `OptimizeContentFunction` in `igny8_core/ai/functions/optimize_content.py`
2. Add cluster-alignment, keyword coverage, heading analysis, intent classification, scoring
3. Maintain backward compatibility — existing `optimize_content` calls still work

### Step 4: API Endpoints
1. Add optimizer endpoints to `igny8_core/urls/optimizer.py` (or create if doesn't exist)
2. Create views: `AnalyzeView`, `OptimizeView`, `PreviewView`, `ApplyView`, `RejectView`, `BatchView`
3. Create `ClusterSuggestionsView`, `AssignClusterView`, `DashboardView`, `DiffView`
4. Register URL patterns under `/api/v1/optimizer/`

### Step 5: Celery Tasks
1. Implement `run_optimization`, `run_batch_optimization`, `identify_optimization_candidates`
2. Add `identify_optimization_candidates` to Celery beat schedule

### Step 6: Serializers & Admin
1. Update DRF serializer for extended OptimizationTask (include all 16 new fields)
2. Create nested serializers for before/after views
3. Update Django admin registration

### Step 7: Credit Cost Configuration
Add to `CreditCostConfig` (billing app):

| operation_type | default_cost | description |
|---------------|-------------|-------------|
| `optimization_analysis` | 2 | Analyze single content (scoring + keyword coverage) |
| `optimization_full_rewrite` | 5-8 | Full rewrite optimization (varies by content length) |
| `optimization_schema_only` | 1 | Schema gap fix only |
| `optimization_batch` | 15-25 | Batch optimization for 10 items |

Credit deduction follows existing `CreditUsageLog` pattern.

---

## 5. ACCEPTANCE CRITERIA

### Cluster Matching
- [ ] Content without cluster assignment gets auto-matched with confidence scoring
- [ ] Confidence ≥ 0.6 auto-assigns; < 0.6 flags for manual review with top 3 suggestions
- [ ] Cluster suggestions endpoint returns ranked candidates

### Keyword Coverage
- [ ] All cluster keywords analyzed for presence in content
- [ ] Coverage report includes exact match, partial match, and missing keywords
- [ ] Hub content targets 70%+, supporting articles 40%+, product/service 30%+

### Heading Restructure
- [ ] H1/H2/H3 hierarchy validated (single H1, no skipped levels)
- [ ] Missing keyword themes identified and new headings suggested
- [ ] AI rewrites headings incorporating target keywords while maintaining meaning

### Content Rewrite
- [ ] Intent classified correctly (informational/commercial/transactional)
- [ ] Rewrite adjusts content structure based on intent
- [ ] Thin content expanded, bloated content compressed
- [ ] Missing keyword sections added

### Scoring
- [ ] Score 0-100 calculated with 5 weighted factors
- [ ] score_before recorded before any changes
- [ ] score_after recorded after optimization
- [ ] Dashboard shows average improvement and distribution

### Before/After
- [ ] Full snapshot of original content preserved in content_before
- [ ] Optimized version stored in content_after without auto-applying
- [ ] Diff view provides visual HTML comparison
- [ ] Apply action copies content_after → Content.content_html
- [ ] Reject action preserves original, marks task rejected

### Batch
- [ ] Batch optimization selects content by cluster, score threshold, type, or explicit IDs
- [ ] Max 3 concurrent optimizations per account enforced
- [ ] Progress trackable via OptimizationTask status
- [ ] Weekly candidate identification runs without auto-optimizing

### Integration
- [ ] Schema gap detection leverages SchemaValidationService from 02G
- [ ] Credit costs deducted per CreditCostConfig entries
- [ ] All API endpoints respect account/site permission boundaries

---

## 6. CLAUDE CODE INSTRUCTIONS

### File Locations
```
igny8_core/
├── ai/
│   └── functions/
│       └── optimize_content.py       # Enhance existing function
├── business/
│   ├── content_scoring.py            # ContentScoringService
│   └── keyword_coverage.py           # KeywordCoverageAnalyzer
├── tasks/
│   └── optimization_tasks.py         # Celery tasks
├── urls/
│   └── optimizer.py                  # Optimizer endpoints
└── migrations/
    └── XXXX_extend_optimization_task.py
```

### Conventions
- **PKs:** BigAutoField (integer) — do NOT use UUIDs
- **Table prefix:** `igny8_` (existing table `igny8_optimization_tasks`)
- **Celery app name:** `igny8_core`
- **URL pattern:** `/api/v1/optimizer/...`
- **Permissions:** Use `SiteSectorModelViewSet` permission pattern
- **AI functions:** Extend existing `BaseAIFunction` subclass — do NOT create a new registration key, enhance the existing `optimize_content`
- **Frontend:** `.tsx` files with Zustand stores for state management

### Cross-References
| Doc | Relationship |
|-----|-------------|
| **02B** | Taxonomy terms get cluster context for optimization; ClusterMappingService scoring pattern reused |
| **02G** | SchemaValidationService used for schema gap detection; schema_only optimization triggers 02G schema generation |
| **02C** | GSC position data identifies pages needing optimization (high impressions, low clicks) |
| **02D** | Optimizer identifies internal link opportunities and feeds them to linker |
| **01E** | Blueprint-aware pipeline sets initial content quality; optimizer improves post-generation |
| **01A** | SAGBlueprint/SAGCluster data provides cluster context for optimization |
| **01G** | SAG health monitoring can incorporate content quality scores as a health factor |

### Key Decisions
1. **Extend, don't replace** — The existing `OptimizationTask` model and `optimize_content` AI function are enhanced, not replaced with new models
2. **Preview-first workflow** — Optimizations always produce a preview (status=`review`) before applying to Content
3. **Content snapshot** — Full HTML snapshot stored in `content_before` for rollback capability
4. **Score reuse** — `ContentScoringService` is a standalone service usable by other modules (02G schema audit, 01G health monitoring)
5. **Schema delegation** — Schema gap detection reuses 02G's `SchemaValidationService` rather than duplicating logic