bakery-ia/services/alert_processor/README.md

# Unified Alert Service

## 🎉 Latest Updates (Waves 3-6 Complete)

### Wave 6: Production-Ready Deployment
- **Database Migration** - Clean break from legacy fields (`severity`, `actions`) → [20251122_1000_remove_legacy_alert_fields.py](migrations/versions/20251122_1000_remove_legacy_alert_fields.py)
- **Backfill Script** - Enriches existing alerts with missing priority scores and smart actions → [backfill_enriched_alerts.py](scripts/backfill_enriched_alerts.py)
- **Integration Tests** - Comprehensive test suite for enriched alert flow → [test_enriched_alert_flow.py](tests/integration/test_enriched_alert_flow.py)
- **API Documentation** - Complete reference guide with examples → [ENRICHED_ALERTS_API.md](docs/ENRICHED_ALERTS_API.md)

### Wave 5: Enhanced UX Features (Frontend)
- **Trend Visualizations** - Inline sparklines for TREND_WARNING alerts
- **Action Consequence Previews** - See outcomes before taking action (financial impact, reversibility)
- **Response Time Gamification** - Track performance metrics by priority level with benchmarks

### Wave 4: High-Priority Features
- **Email Digest Service** - Celebration-first daily/weekly summaries → [email_digest.py](app/services/enrichment/email_digest.py)
- **Email Digest API** - POST `/api/v1/tenants/{id}/alerts/digest/send`
- **Alert Hub** (Frontend) - 3-tab organization (All/For Me/Archived)
- **Auto-Action Countdown** (Frontend) - Real-time timer with one-click cancel
- **Priority Score Explainer** (Frontend) - Educational transparency modal

### Wave 3: Dashboard Widgets (Frontend)
- **AI Handling Rate Card** - Showcase AI wins (handling %, savings EUR, trend)
- **Prevented Issues Card** - Celebrate problems AI automatically resolved

## Overview

The **Unified Alert Service** (formerly alert_processor) is the intelligent, centralized alert system for the Bakery-IA platform. It automatically enriches all alerts with multi-factor priority scoring, orchestrator context (AI actions), business impact analysis, **context-aware message generation**, smart actions with deep links, and timing intelligence. This service combines real-time alert processing, intelligent enrichment, and multi-channel delivery into a single, powerful microservice.

**Key Innovation**: Unlike traditional alert systems that simply route messages, the Unified Alert Service applies sophisticated AI-driven enrichment to transform raw alerts into actionable insights. Every alert is automatically analyzed for business impact, enriched with context from AI orchestrator decisions, scored using multi-factor algorithms, **generates intelligent messages using enrichment data**, and enhanced with smart actions that include deep links and one-click solutions.

## Architecture Evolution

### Unified Service (Current)
**Single Service**: `alert-processor` with integrated enrichment
- ✅ All alerts automatically enriched with priority, context, and actions
- ✅ Multi-factor priority scoring (business impact, urgency, user agency, confidence)
- ✅ Orchestrator context injection (AI already handled)
- ✅ Smart actions with deep links and phone dialing
- ✅ Timing intelligence for optimal delivery
- ✅ Single source of truth for all alert data
- ✅ 50% less infrastructure complexity

**Benefits**:
- **Automatic Enrichment**: Every alert enriched by default, no manual configuration
- **Better Data Consistency**: Single database, single service, single source of truth
- **Simpler Operations**: One service to monitor, deploy, and scale
- **Faster Development**: Changes to enrichment logic in one place
- **Lower Latency**: No inter-service communication for enrichment

## Key Features

### 🎯 Multi-Factor Priority Scoring

Sophisticated priority algorithm that scores alerts 0-100 using weighted factors:

**Priority Weights** (configurable):
- **Business Impact** (40%): Financial impact, affected orders, customer impact
- **Urgency** (30%): Time sensitivity, deadlines, production dependencies
- **User Agency** (20%): Whether user can/must take action
- **Confidence** (10%): AI confidence in the analysis

**Priority Levels**:
- **Critical (90-100)**: Immediate action required → WhatsApp + Email + Push
- **Important (70-89)**: Action needed soon → WhatsApp + Email (business hours)
- **Standard (50-69)**: Should be addressed → Email (business hours)
- **Info (0-49)**: For awareness → Dashboard only

**Example**: A supplier delay affecting tomorrow's production gets scored:
- Business Impact: High (€450 lost revenue) = 90/100
- Urgency: Critical (6 hours until production) = 95/100
- User Agency: Must act (call supplier) = 100/100
- Confidence: High (clear data) = 92/100
- **Final Score**: 92 → **CRITICAL** → WhatsApp immediately

### 🧠 Orchestrator Context Enrichment

Enriches alerts with AI orchestrator actions to provide "AI already handled" context:

**Context Types**:
- **Prevented Issues**: AI created purchase order before stockout
- **Weather-Based Actions**: AI adjusted production for sunny weekend
- **Waste Reduction**: AI optimized production based on patterns
- **Proactive Ordering**: AI detected trend and ordered supplies

**Example Alert Enrichment**:
```
Alert: "Low Stock: Harina Tipo 55 (45kg, Min: 200kg)"
Orchestrator Context:
  ✓ AI already handled: Purchase order #12345 created 2 hours ago
  ✓ 500kg arriving Friday (2.3 days before predicted stockout)
  ✓ Estimated savings: €200 (prevented stockout)
  ✓ Confidence: 92%
Result: Priority reduced from CRITICAL (85) to IMPORTANT (71)
Type: "Prevented Issue" (not "Action Needed")
```

### 🎨 Smart Actions with Deep Links

Enhanced actions with metadata for one-click execution:

**Smart Action Features**:
- **Deep Links**: Direct navigation to relevant pages
  - `action://inventory/item/{ingredient_id}` → Inventory detail page
  - `action://procurement/create-po?ingredient={id}` → Pre-filled PO form
  - `action://production/batch/{batch_id}` → Production batch view
- **Phone Dialing**: `tel:+34-555-1234` for immediate calls
- **Email Links**: `mailto:supplier@example.com` with pre-filled subjects
- **URL Parameters**: Pre-populate forms with context
- **Action Metadata**: Additional data for client-side processing

**Example Smart Actions**:
```json
{
  "smart_actions": [
    {
      "label": "Call Supplier Now",
      "type": "phone",
      "url": "tel:+34-555-1234",
      "metadata": {
        "supplier_name": "Levadura Fresh",
        "contact_name": "Juan García"
      }
    },
    {
      "label": "View Purchase Order",
      "type": "navigation",
      "url": "action://procurement/po/12345",
      "metadata": {
        "po_number": "PO-12345",
        "status": "pending_delivery"
      }
    },
    {
      "label": "View Ingredient Details",
      "type": "navigation",
      "url": "action://inventory/item/flour-tipo-55",
      "metadata": {
        "current_stock": 45,
        "min_stock": 200
      }
    }
  ]
}
```

### ⏰ Timing Intelligence

Optimizes alert delivery based on business hours and user attention patterns:

**Timing Decisions**:
- **SEND_NOW**: Critical alerts, immediate action required
- **SCHEDULE_LATER**: Important alerts sent during peak attention hours
- **BATCH_FOR_DIGEST**: Low-priority alerts batched for end-of-day digest

**Peak Hours Detection**:
- **Morning Peak**: 7:00-11:00 (high attention, good for planning)
- **Evening Peak**: 17:00-19:00 (transition time, good for summaries)
- **Quiet Hours**: 22:00-6:00 (only critical alerts)

**Business Hours**: 6:00-22:00 (configurable per tenant)

**Example**:
```
Alert Type: Waste Trend (Standard priority, 58/100)
Current Time: 23:30 (quiet hours)
Decision: BATCH_FOR_DIGEST (send in 18:00 digest tomorrow)
Reasoning: Non-urgent, better addressed during business hours
```

### 🏷️ Alert Type Classification

Categorizes alerts by user job-to-be-done (JTBD):

**Type Classes**:
1. **ACTION_NEEDED**: User must take action (call supplier, create PO)
2. **PREVENTED_ISSUE**: AI already handled, FYI only (PO created automatically)
3. **TREND_WARNING**: Pattern detected, consider adjusting (waste increasing)
4. **ESCALATION**: Previous alert ignored, now more urgent
5. **INFORMATION**: For awareness only (forecast updated)

**UI/UX Benefits**:
- Different icons per type class
- Color coding (red = action, green = prevented, yellow = trend)
- Smart filtering by type
- Different notification sounds

### 📊 Business Impact Analysis

Calculates and displays financial impact:

**Impact Factors**:
- **Revenue Impact**: Lost sales, affected orders
- **Cost Impact**: Waste, expedited shipping, overtime
- **Customer Impact**: Affected orders, potential cancellations
- **Operational Impact**: Production delays, equipment downtime

**Example**:
```json
{
  "business_impact": {
    "financial_impact_eur": 450,
    "affected_orders": 3,
    "affected_products": ["Croissant Mantequilla"],
    "production_delay_hours": 6,
    "estimated_revenue_loss_eur": 450,
    "impact_level": "high"
  }
}
```

### 🔄 Real-Time SSE Streaming

Server-Sent Events for instant dashboard updates:

**Features**:
- Real-time alert delivery to connected dashboards
- Enriched alerts streamed immediately after processing
- Per-tenant channels (`alerts:{tenant_id}`)
- Automatic reconnection handling
- Cached active alerts for new connections

**Client Integration**:
```javascript
const eventSource = new EventSource(`/api/v1/alerts/stream?tenant_id=${tenantId}`);

eventSource.onmessage = (event) => {
  const enrichedAlert = JSON.parse(event.data);

  console.log('Priority:', enrichedAlert.priority_score);
  console.log('Type:', enrichedAlert.type_class);
  console.log('AI Handled:', enrichedAlert.orchestrator_context?.ai_already_handled);
  console.log('Smart Actions:', enrichedAlert.smart_actions);

  // Update dashboard UI
  displayAlert(enrichedAlert);
};
```

### 🌐 Context-Aware Message Generation with i18n Support

**NEW**: Messages are now generated AFTER enrichment, leveraging all context data for intelligent, multilingual notifications.

**Before** (Static Templates):
```
"Solo 5kg disponibles, necesarios 30kg para producción de mañana"
```
- ❌ Hardcoded "mañana" (may not be tomorrow)
- ❌ No AI action context
- ❌ Missing supplier details
- ❌ Spanish only

**After** (Context-Aware with i18n):
```json
{
  "title": "🚨 Stock Crítico: Flour",
  "message": "Solo 5.0kg de Flour (necesitas 30.0kg el lunes 24). Ya creé PO-12345 con Mill Co para entrega mañana 10:00. Por favor aprueba €150.",
  "metadata": {
    "i18n": {
      "title_key": "alerts.critical_stock_shortage.title",
      "title_params": {"ingredient_name": "Flour"},
      "message_key": "alerts.critical_stock_shortage.message_with_po_pending",
      "message_params": {
        "ingredient_name": "Flour",
        "current_stock": 5.0,
        "required_stock": 30.0,
        "po_id": "PO-12345",
        "delivery_date": "2025-11-24",
        "po_amount": 150.0
      }
    }
  }
}
```
- ✅ Actual date (Monday 24th)
- ✅ AI action mentioned (PO-12345)
- ✅ Supplier and delivery details
- ✅ i18n keys for any language

**Message Variants Based on Context**:
The same stock shortage alert generates different messages based on enrichment:

1. **AI Already Created PO (Pending Approval)**:
   ```
   "Solo 5.0kg de Flour (necesitas 30.0kg el lunes 24).
   Ya creé PO-12345 con Mill Co para entrega mañana 10:00.
   Por favor aprueba €150."
   ```

2. **Supplier Contact Available**:
   ```
   "Solo 5.0kg de Flour (necesitas 30.0kg en 18 horas).
   Contacta a Mill Co (+34 123 456 789)."
   ```

3. **Generic (No Special Context)**:
   ```
   "Solo 5.0kg de Flour disponibles (necesitas 30.0kg)."
   ```

**i18n Integration**:
Frontend uses i18n keys for translation:
```typescript
// English
t("alerts.critical_stock_shortage.message_with_po_pending", params)
// → "Only 5.0kg of Flour (need 30.0kg on Monday 24th). Already created PO-12345..."

// Euskera (Basque)
t("alerts.critical_stock_shortage.message_with_po_pending", params)
// → "Flour-en 5.0kg bakarrik (30.0kg behar dituzu astelehen 24an). PO-12345 sortu dut..."

// Spanish (Fallback)
alert.message
// → "Solo 5.0kg de Flour..."
```

**Template System**:
See [shared/alerts/context_templates.py](../../shared/alerts/context_templates.py) for the complete generator system.

Each alert type has a function that:
1. Accepts `EnrichedAlert` with full context
2. Determines message variant based on:
   - Orchestrator context (AI acted?)
   - Urgency (time-sensitive?)
   - User agency (supplier contact available?)
3. Returns i18n keys, parameters, and fallback messages

## Technology Stack

### Core Technologies
- **Framework**: FastAPI (Python 3.11+) - Async web framework
- **Database**: PostgreSQL 17 - Alert storage with JSONB for flexible enrichment data
- **Caching**: Redis 7.4 - Active alerts cache, SSE pub/sub
- **Messaging**: RabbitMQ 4.1 - Event consumption from all services
- **ORM**: SQLAlchemy 2.0 (async) - Database abstraction
- **Migrations**: Alembic - Database schema management
- **Logging**: Structlog - Structured JSON logging
- **Metrics**: Prometheus Client - Alert metrics and performance

### Enrichment Stack
- **HTTP Client**: httpx (async) - Service-to-service communication
- **Priority Scoring**: Custom multi-factor algorithm
- **Context Enrichment**: Orchestrator API client
- **Timing Intelligence**: Peak hours detection engine
- **Smart Actions**: Deep link generator with metadata

## Database Schema

### Main Tables

#### alerts (Primary Table)
```sql
CREATE TABLE alerts (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    tenant_id UUID NOT NULL,
    item_type VARCHAR(50) NOT NULL,              -- 'alert' or 'recommendation'
    alert_type VARCHAR(100) NOT NULL,            -- low_stock, supplier_delay, waste_trend
    service VARCHAR(100) NOT NULL,               -- inventory, procurement, production

    -- Legacy severity (kept for backward compatibility)
    severity VARCHAR(50) NOT NULL,               -- urgent, high, medium, low

    -- Basic content
    title VARCHAR(500) NOT NULL,
    message TEXT NOT NULL,
    alert_metadata JSONB,                        -- Legacy metadata

    -- Lifecycle
    status VARCHAR(50) DEFAULT 'active',         -- active, acknowledged, resolved
    created_at TIMESTAMP DEFAULT NOW(),

    -- ============================================================
    -- ENRICHMENT FIELDS (NEW - Unified Alert Service)
    -- ============================================================

    -- Multi-factor priority scoring
    priority_score INTEGER NOT NULL CHECK (priority_score >= 0 AND priority_score <= 100),
    priority_level VARCHAR(50) NOT NULL CHECK (priority_level IN ('critical', 'important', 'standard', 'info')),
    type_class VARCHAR(50) NOT NULL CHECK (type_class IN ('action_needed', 'prevented_issue', 'trend_warning', 'escalation', 'information')),

    -- Context enrichment (JSONB for flexibility)
    orchestrator_context JSONB,                  -- AI actions that provide context
    business_impact JSONB,                       -- Financial and operational impact
    urgency_context JSONB,                       -- Time sensitivity details
    user_agency JSONB,                           -- What user can/must do
    trend_context JSONB,                         -- Historical patterns

    -- Smart actions (enhanced with deep links and metadata)
    smart_actions JSONB NOT NULL DEFAULT '[]',   -- Actions with URLs and metadata

    -- AI reasoning
    ai_reasoning_summary TEXT,                   -- Explanation of priority/classification
    confidence_score FLOAT NOT NULL DEFAULT 0.8, -- AI confidence in analysis

    -- Timing intelligence
    timing_decision VARCHAR(50) NOT NULL DEFAULT 'send_now' CHECK (timing_decision IN ('send_now', 'schedule_later', 'batch_for_digest')),
    scheduled_send_time TIMESTAMP,               -- When to send if scheduled

    -- Placement hints
    placement JSONB NOT NULL DEFAULT '["dashboard"]', -- Where to show alert

    -- Performance indexes
    INDEX idx_alerts_tenant_status (tenant_id, status),
    INDEX idx_alerts_priority_score (tenant_id, priority_score DESC, created_at DESC),
    INDEX idx_alerts_type_class (tenant_id, type_class, status),
    INDEX idx_alerts_priority_level (priority_level, status),
    INDEX idx_alerts_timing (timing_decision, scheduled_send_time),
    INDEX idx_alerts_created (tenant_id, created_at DESC)
);
```

**Key Design Decisions**:
- **JSONB for Context**: Flexible structure for enrichment data
- **NOT NULL Defaults**: All alerts must have enrichment fields
- **Check Constraints**: Data integrity for enums and ranges
- **Indexes**: Optimized for dashboard queries (priority, date)
- **No Legacy Support**: Clean schema, no migration artifacts

### Enrichment Data Structures

#### orchestrator_context (JSONB)
```json
{
  "ai_already_handled": true,
  "action_type": "purchase_order_created",
  "action_id": "uuid",
  "action_summary": "Created PO #12345 for 500kg flour",
  "reasoning": "Detected stockout risk 2.3 days ahead",
  "estimated_savings_eur": 200,
  "prevented_issue": "stockout",
  "confidence": 0.92,
  "created_at": "2025-11-21T10:00:00Z"
}
```

#### business_impact (JSONB)
```json
{
  "financial_impact_eur": 450,
  "affected_orders": 3,
  "affected_products": ["Croissant Mantequilla"],
  "production_delay_hours": 6,
  "estimated_revenue_loss_eur": 450,
  "customer_impact": "high",
  "impact_level": "high"
}
```

#### urgency_context (JSONB)
```json
{
  "deadline": "2025-11-22T08:00:00Z",
  "time_until_deadline_hours": 6,
  "dependencies": ["production_batch_croissants_001"],
  "urgency_level": "high",
  "reason": "Production starts in 6 hours"
}
```

#### user_agency (JSONB)
```json
{
  "can_act": true,
  "must_act": true,
  "action_type": "call_supplier",
  "action_urgency": "immediate",
  "alternative_actions": ["find_alternative_supplier", "delay_production"]
}
```

#### smart_actions (JSONB Array)
```json
[
  {
    "label": "Call Supplier Now",
    "type": "phone",
    "url": "tel:+34-555-1234",
    "metadata": {
      "supplier_name": "Levadura Fresh",
      "contact_name": "Juan García",
      "contact_role": "Sales Manager"
    }
  },
  {
    "label": "View Purchase Order",
    "type": "navigation",
    "url": "action://procurement/po/12345",
    "metadata": {
      "po_number": "PO-12345",
      "status": "pending_delivery",
      "estimated_delivery": "2025-11-22T14:00:00Z"
    }
  },
  {
    "label": "Email Supplier",
    "type": "email",
    "url": "mailto:pedidos@levadura-fresh.es?subject=Urgent:%20Order%20Delay",
    "metadata": {
      "template": "supplier_delay_urgent"
    }
  }
]
```

## API Endpoints

### Alert Management

#### GET /api/v1/alerts
List alerts with filtering and enrichment data.

**Query Parameters**:
- `tenant_id` (required): UUID
- `status`: active, acknowledged, resolved
- `priority_level`: critical, important, standard, info
- `type_class`: action_needed, prevented_issue, etc.
- `min_priority_score`: 0-100
- `limit`: Default 50, max 500
- `offset`: Pagination offset

**Response** (enriched):
```json
{
  "alerts": [
    {
      "id": "uuid",
      "tenant_id": "uuid",
      "item_type": "alert",
      "alert_type": "supplier_delay",
      "service": "procurement",
      "title": "Supplier Delay: Levadura Fresh",
      "message": "Delivery delayed 24 hours",

      "priority_score": 92,
      "priority_level": "critical",
      "type_class": "action_needed",

      "orchestrator_context": {
        "ai_already_handled": false
      },
      "business_impact": {
        "financial_impact_eur": 450,
        "affected_orders": 3
      },
      "smart_actions": [
        {
          "label": "Call Supplier Now",
          "type": "phone",
          "url": "tel:+34-555-1234"
        }
      ],
      "ai_reasoning_summary": "Critical priority due to production deadline in 6 hours and €450 impact",
      "confidence_score": 0.92,

      "timing_decision": "send_now",
      "placement": ["dashboard", "notifications"],

      "created_at": "2025-11-21T10:00:00Z",
      "status": "active"
    }
  ],
  "total": 42,
  "page": 1,
  "pages": 1
}
```

#### GET /api/v1/alerts/stream (SSE)
Server-Sent Events stream for real-time enriched alerts.

**Query Parameters**:
- `tenant_id` (required): UUID

**Stream Format**:
```
data: {"id": "uuid", "priority_score": 92, "type_class": "action_needed", ...}

data: {"id": "uuid2", "priority_score": 58, "type_class": "trend_warning", ...}
```

#### POST /api/v1/alerts/{alert_id}/acknowledge
Acknowledge alert (calculates response time).

**Request Body**:
```json
{
  "user_id": "uuid",
  "notes": "Called supplier, delivery confirmed for tomorrow"
}
```

#### POST /api/v1/alerts/{alert_id}/resolve
Mark alert as resolved (calculates resolution time).

**Request Body**:
```json
{
  "user_id": "uuid",
  "resolution_notes": "Issue resolved, delivery received",
  "outcome": "successful"
}
```

### Alert Analytics (Enriched)

#### GET /api/v1/alerts/analytics/dashboard
Dashboard metrics with enrichment insights.

**Response**:
```json
{
  "summary": {
    "total_alerts": 150,
    "critical_alerts": 12,
    "important_alerts": 38,
    "standard_alerts": 75,
    "info_alerts": 25,
    "avg_priority_score": 65,
    "ai_handled_percentage": 35,
    "action_needed_percentage": 45,
    "prevented_issues_count": 52
  },
  "by_priority_level": {
    "critical": 12,
    "important": 38,
    "standard": 75,
    "info": 25
  },
  "by_type_class": {
    "action_needed": 68,
    "prevented_issue": 52,
    "trend_warning": 20,
    "escalation": 5,
    "information": 5
  },
  "business_impact_total_eur": 12500,
  "avg_response_time_minutes": 8,
  "avg_resolution_time_minutes": 45
}
```

### Health & Monitoring

#### GET /health
Service health check with enrichment service status.

**Response**:
```json
{
  "status": "healthy",
  "database": "connected",
  "redis": "connected",
  "rabbitmq": "connected",
  "enrichment_services": {
    "priority_scoring": "operational",
    "context_enrichment": "operational",
    "timing_intelligence": "operational",
    "orchestrator_client": "connected"
  },
  "metrics": {
    "items_processed": 1250,
    "items_stored": 1250,
    "enrichments_count": 1250,
    "enrichment_success_rate": 0.98,
    "avg_enrichment_time_ms": 45
  }
}
```

## Enrichment Pipeline

### Processing Flow

```
1. RabbitMQ Event → Raw Alert
   ↓
2. Enrich Alert (automatic)
   - Query orchestrator for AI actions
   - Calculate multi-factor priority score
   - Classify alert type (action_needed vs prevented_issue)
   - Analyze business impact
   - Assess urgency and user agency
   - Generate smart actions with deep links
   - Apply timing intelligence
   - Add AI reasoning summary
   ↓
3. Store Enriched Alert
   - Save to PostgreSQL with all enrichment fields
   - Cache in Redis for SSE
   ↓
4. Route by Priority Score
   - Critical (90-100): WhatsApp + Email + Push
   - Important (70-89): WhatsApp + Email (business hours)
   - Standard (50-69): Email (business hours)
   - Info (0-49): Dashboard only
   ↓
5. Stream to SSE
   - Publish enriched alert to Redis
   - Dashboard receives real-time update
```

### Enrichment Services

#### 1. Priority Scoring Service
**File**: `app/services/enrichment/priority_scoring.py`

Calculates priority score using multi-factor algorithm:

```python
def calculate_priority_score(alert: dict, context: dict) -> int:
    """
    Calculate 0-100 priority score using weighted factors.

    Weights:
    - Business Impact: 40%
    - Urgency: 30%
    - User Agency: 20%
    - Confidence: 10%
    """
    business_score = calculate_business_impact_score(alert)  # 0-100
    urgency_score = calculate_urgency_score(alert)           # 0-100
    agency_score = calculate_user_agency_score(alert)        # 0-100
    confidence = context.get('confidence', 0.8)              # 0-1

    priority = (
        business_score * 0.4 +
        urgency_score * 0.3 +
        agency_score * 0.2 +
        (confidence * 100) * 0.1
    )

    return int(priority)

def determine_priority_level(score: int) -> str:
    """Map score to priority level."""
    if score >= 90:
        return 'critical'
    elif score >= 70:
        return 'important'
    elif score >= 50:
        return 'standard'
    else:
        return 'info'
```

#### 2. Context Enrichment Service
**File**: `app/services/enrichment/context_enrichment.py`

Queries orchestrator for AI actions and enriches alert context:

```python
async def enrich_alert(self, alert: dict) -> dict:
    """
    Main enrichment orchestrator.
    Coordinates all enrichment services.
    """
    # Query orchestrator for AI actions
    orchestrator_context = await self.get_orchestrator_context(alert)

    # Calculate business impact
    business_impact = self.calculate_business_impact(alert, orchestrator_context)

    # Assess urgency
    urgency_context = self.assess_urgency(alert)

    # Evaluate user agency
    user_agency = self.evaluate_user_agency(alert, orchestrator_context)

    # Calculate priority score
    priority_score = self.priority_scoring.calculate_priority(
        alert,
        business_impact,
        urgency_context,
        user_agency
    )

    # Classify alert type
    type_class = self.classify_alert_type(alert, orchestrator_context)

    # Generate smart actions
    smart_actions = self.generate_smart_actions(alert, orchestrator_context)

    # Apply timing intelligence
    timing_decision = self.timing_intelligence.determine_timing(
        priority_score,
        datetime.now().hour
    )

    # Generate AI reasoning
    ai_reasoning = self.generate_reasoning(
        priority_score,
        type_class,
        business_impact,
        orchestrator_context
    )

    return {
        **alert,
        'priority_score': priority_score,
        'priority_level': self.determine_priority_level(priority_score),
        'type_class': type_class,
        'orchestrator_context': orchestrator_context,
        'business_impact': business_impact,
        'urgency_context': urgency_context,
        'user_agency': user_agency,
        'smart_actions': smart_actions,
        'ai_reasoning_summary': ai_reasoning,
        'confidence_score': orchestrator_context.get('confidence', 0.8),
        'timing_decision': timing_decision,
        'placement': self.determine_placement(priority_score, type_class)
    }
```

#### 3. Orchestrator Client
**File**: `app/services/enrichment/orchestrator_client.py`

HTTP client for querying orchestrator service:

```python
class OrchestratorClient:
    """Client for querying orchestrator service for AI actions."""

    async def get_recent_actions(
        self,
        tenant_id: str,
        ingredient_id: Optional[str] = None,
        hours_ago: int = 24
    ) -> List[Dict[str, Any]]:
        """
        Query orchestrator for recent AI actions.
        Used to determine if AI already handled the issue.
        """
        try:
            response = await self.client.get(
                f"{self.base_url}/api/internal/recent-actions",
                params={
                    'tenant_id': tenant_id,
                    'ingredient_id': ingredient_id,
                    'hours_ago': hours_ago
                },
                headers={'X-Internal-Service': 'alert-processor'}
            )

            if response.status_code == 200:
                return response.json().get('actions', [])
            return []

        except Exception as e:
            logger.error("Failed to query orchestrator", error=str(e))
            return []
```

#### 4. Timing Intelligence Service
**File**: `app/services/enrichment/timing_intelligence.py`

Determines optimal alert delivery timing:

```python
class TimingIntelligenceService:
    """Determines optimal alert delivery timing."""

    def determine_timing(self, priority_score: int, current_hour: int) -> str:
        """
        Determine when to send alert based on priority and time.

        Returns:
            'send_now': Send immediately
            'schedule_later': Schedule for peak hours
            'batch_for_digest': Add to end-of-day digest
        """
        # Critical always sends now
        if priority_score >= 90:
            return 'send_now'

        # Important sends during business hours
        if priority_score >= 70:
            if self.is_business_hours(current_hour):
                return 'send_now'
            else:
                return 'schedule_later'  # Send at business hours start

        # Standard batches during quiet hours
        if priority_score >= 50:
            if self.is_peak_hours(current_hour):
                return 'send_now'
            else:
                return 'batch_for_digest'

        # Info always batches
        return 'batch_for_digest'

    def is_peak_hours(self, hour: int) -> bool:
        """Check if current hour is peak attention time."""
        morning_peak = 7 <= hour <= 11
        evening_peak = 17 <= hour <= 19
        return morning_peak or evening_peak

    def is_business_hours(self, hour: int) -> bool:
        """Check if current hour is business hours."""
        return 6 <= hour <= 22
```

## Configuration

### Environment Variables

**Service Configuration:**
```bash
# Service
PORT=8010
SERVICE_NAME=alert-processor

# Database
ALERT_PROCESSOR_DB_USER=alert_processor_user
ALERT_PROCESSOR_DB_PASSWORD=<secure_password>
ALERT_PROCESSOR_DB_HOST=alert-processor-db-service
ALERT_PROCESSOR_DB_PORT=5432
ALERT_PROCESSOR_DB_NAME=alert_processor_db

# Redis
REDIS_URL=rediss://redis-service:6379/6
REDIS_MAX_CONNECTIONS=50

# RabbitMQ
RABBITMQ_URL=amqp://user:pass@rabbitmq-service:5672/

# Alert Processing
ALERT_BATCH_SIZE=10
ALERT_PROCESSING_TIMEOUT=30
ALERT_DEDUPLICATION_WINDOW_MINUTES=15
```

**Enrichment Configuration:**
```bash
# Priority Scoring Weights (must sum to 1.0)
BUSINESS_IMPACT_WEIGHT=0.4
URGENCY_WEIGHT=0.3
USER_AGENCY_WEIGHT=0.2
CONFIDENCE_WEIGHT=0.1

# Priority Thresholds (0-100 scale)
CRITICAL_THRESHOLD=90
IMPORTANT_THRESHOLD=70
STANDARD_THRESHOLD=50

# Timing Intelligence
BUSINESS_HOURS_START=6
BUSINESS_HOURS_END=22
PEAK_HOURS_START=7
PEAK_HOURS_END=11
PEAK_HOURS_EVENING_START=17
PEAK_HOURS_EVENING_END=19

# Alert Grouping
GROUPING_TIME_WINDOW_MINUTES=15
MAX_ALERTS_PER_GROUP=5

# Email Digest
DIGEST_SEND_TIME=18:00

# Service URLs for Enrichment
ORCHESTRATOR_SERVICE_URL=http://orchestrator-service:8000
INVENTORY_SERVICE_URL=http://inventory-service:8000
PRODUCTION_SERVICE_URL=http://production-service:8000
```

## Deployment

### Prerequisites
- PostgreSQL 17 with alert_processor database
- Redis 7.4
- RabbitMQ 4.1
- Python 3.11+

### Database Migration

```bash
cd services/alert_processor

# Run migration to add enrichment fields
alembic upgrade head

# Verify migration
psql $DATABASE_URL -c "\d alerts"
# Should show all enrichment columns:
# - priority_score, priority_level, type_class
# - orchestrator_context, business_impact, smart_actions
# - ai_reasoning_summary, confidence_score
# - timing_decision, scheduled_send_time, placement
```

### Kubernetes Deployment

```bash
# Deploy with Tilt (development)
tilt up

# Or deploy with kubectl
kubectl apply -k infrastructure/kubernetes/overlays/dev

# Verify deployment
kubectl get pods -l app.kubernetes.io/name=alert-processor-service -n bakery-ia

# Check enrichment logs
kubectl logs -f deployment/alert-processor-service -n bakery-ia | grep "enriched"
```

### Verify Enrichment

```bash
# Seed demo alerts
python services/demo_session/scripts/seed_enriched_alert_demo.py

# Check database for enriched fields
psql $DATABASE_URL -c "
  SELECT
    id,
    title,
    priority_score,
    priority_level,
    type_class,
    orchestrator_context->>'ai_already_handled' as ai_handled,
    jsonb_array_length(smart_actions) as action_count
  FROM alerts
  ORDER BY created_at DESC
  LIMIT 5;
"

# Should see alerts with:
# - Priority scores (0-100)
# - Priority levels (critical, important, standard, info)
# - Type classes (action_needed, prevented_issue, etc.)
# - Orchestrator context (if AI handled)
# - Smart actions (multiple actions per alert)
```

## Demo Data

### Seed Enriched Demo Alerts

```bash
# Run demo seed script
python services/demo_session/scripts/seed_enriched_alert_demo.py
```

**Demo Alerts Created**:

1. **Low Stock Alert** (Important - Prevented Issue)
   - Priority: 71 (AI already handled)
   - Type: prevented_issue
   - Context: AI created purchase order 2 hours ago
   - Smart Actions: View PO, View Ingredient, Call Supplier

2. **Supplier Delay** (Critical - Action Needed)
   - Priority: 92 (6 hours until production)
   - Type: action_needed
   - Impact: €450, 3 affected orders
   - Smart Actions: Call Supplier, Email Supplier, View Batch

3. **Waste Trend** (Standard - Trend Warning)
   - Priority: 58 (pattern detected)
   - Type: trend_warning
   - Pattern: Wednesday overproduction
   - Smart Actions: View Analytics, Adjust Production

4. **Forecast Update** (Info - Information)
   - Priority: 35 (for awareness)
   - Type: information
   - Context: Weather-based demand increase
   - Smart Actions: View Forecast, View Production Plan

5. **Equipment Maintenance** (Standard - Action Needed)
   - Priority: 65 (scheduled maintenance)
   - Type: action_needed
   - Timeline: 48 hours
   - Smart Actions: Schedule Maintenance, Call Technician

## Business Value

### Problem Statement
Spanish bakeries face:
- **Alert Overload**: Too many notifications, no prioritization
- **Missed Critical Issues**: Important alerts buried in noise
- **Lack of Context**: Alerts don't explain why they matter
- **No Action Guidance**: Users don't know what to do next
- **Poor Timing**: Alerts sent at inconvenient times

### Solution: Intelligent Enrichment

**The Unified Alert Service transforms raw alerts into actionable insights:**

1. **Multi-Factor Priority Scoring** → Know what matters most
2. **Orchestrator Context** → See what AI already handled
3. **Business Impact** → Understand financial consequences
4. **Smart Actions** → One-click solutions with deep links
5. **Timing Intelligence** → Receive alerts when you can act

### Quantifiable Impact

**Issue Detection:**
- 90% faster detection (real-time vs. hours/days)
- 50-80% downtime reduction through early warning
- €500-2,000/month cost avoidance (prevented issues)

**Operational Efficiency:**
- 70% fewer false alarms (intelligent filtering)
- 60% faster resolution (smart actions + deep links)
- 2-4 hours/week saved (no manual investigation)
- 85% of alerts include AI reasoning

**Alert Quality:**
- 90%+ alerts are actionable (vs. 30-50% without enrichment)
- 95%+ critical alerts acknowledged within SLA
- 35% of alerts show "AI already handled" (reduced workload)
- 100% of alerts include business impact and smart actions

### ROI Calculation

**Monthly Value per Bakery:**
- Cost Avoidance: €500-2,000 (prevented stockouts, quality issues)
- Time Savings: 2-4 hours/week × €15/hour = €120-240
- Faster Response: 60% improvement = €100-300 value
- **Total Monthly Value**: €720-2,540

**Annual ROI**: €8,640-30,480 value per bakery

**Investment**: €0 additional (included in subscription)

**Payback**: Immediate

## Integration Points

### Dependencies
- **All Services** - Consumes alert events via RabbitMQ
- **Orchestrator Service** - Queries for AI actions and context
- **Inventory Service** - Ingredient and stock data
- **Production Service** - Production batch data
- **Notification Service** - Multi-channel delivery
- **PostgreSQL** - Alert storage
- **Redis** - SSE pub/sub, active alerts cache
- **RabbitMQ** - Event consumption

### Dependents
- **Frontend Dashboard** - Real-time alert display
- **Mobile App** - WhatsApp/Push notifications
- **Analytics Service** - Alert metrics and trends

## Development

### Local Setup

```bash
cd services/alert_processor

# Create virtual environment
python -m venv venv
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Set environment variables
export DATABASE_URL=postgresql+asyncpg://user:pass@localhost:5432/alert_processor_db
export REDIS_URL=redis://localhost:6379/6
export RABBITMQ_URL=amqp://guest:guest@localhost:5672/
export ORCHESTRATOR_SERVICE_URL=http://localhost:8000

# Run migrations
alembic upgrade head

# Start service
python -m app.main
```

### Testing Enrichment

```bash
# Start service
python -m app.main

# In another terminal, send test alert
python -c "
import asyncio
from shared.messaging.rabbitmq import RabbitMQClient

async def test():
    client = RabbitMQClient('amqp://guest:guest@localhost:5672/', 'test')
    await client.connect()
    await client.declare_exchange('alerts', 'topic', durable=True)
    await client.publish('alerts', 'alert.inventory.low_stock', {
        'id': 'test-123',
        'tenant_id': 'demo-tenant-bakery-ia',
        'item_type': 'alert',
        'service': 'inventory',
        'type': 'low_stock',
        'severity': 'warning',
        'title': 'Test: Low Stock',
        'message': 'Stock: 50kg',
        'metadata': {'ingredient_id': 'flour-tipo-55'},
        'actions': [],
        'timestamp': '2025-11-21T10:00:00Z'
    })
    await client.disconnect()

asyncio.run(test())
"

# Check logs for enrichment
# Should see:
# - "Alert enriched successfully"
# - priority_score, priority_level, type_class
# - Orchestrator context queried
# - Smart actions generated
```

## Monitoring

### Metrics (Prometheus)

```python
# Alert processing
alerts_processed_total = Counter(
    'alerts_processed_total',
    'Total alerts processed',
    ['tenant_id', 'service', 'alert_type']
)

# Enrichment
alerts_enriched_total = Counter(
    'alerts_enriched_total',
    'Total alerts enriched',
    ['tenant_id', 'priority_level', 'type_class']
)

enrichment_duration_seconds = Histogram(
    'enrichment_duration_seconds',
    'Time to enrich alert',
    ['enrichment_type'],
    buckets=[0.01, 0.05, 0.1, 0.5, 1.0, 2.0, 5.0]
)

# Priority distribution
alerts_by_priority = Gauge(
    'alerts_by_priority',
    'Active alerts by priority level',
    ['tenant_id', 'priority_level']
)

# Type classification
alerts_by_type_class = Gauge(
    'alerts_by_type_class',
    'Active alerts by type class',
    ['tenant_id', 'type_class']
)

# Orchestrator context
alerts_with_orchestrator_context = Gauge(
    'alerts_with_orchestrator_context',
    'Alerts enriched with orchestrator context',
    ['tenant_id']
)

# Smart actions
avg_smart_actions_per_alert = Gauge(
    'avg_smart_actions_per_alert',
    'Average number of smart actions per alert',
    ['tenant_id']
)
```

### Health Check

```bash
# Check service health
curl http://localhost:8010/health

# Expected response:
{
  "status": "healthy",
  "database": "connected",
  "redis": "connected",
  "rabbitmq": "connected",
  "enrichment_services": {
    "priority_scoring": "operational",
    "context_enrichment": "operational",
    "timing_intelligence": "operational",
    "orchestrator_client": "connected"
  },
  "metrics": {
    "items_processed": 1250,
    "enrichments_count": 1250,
    "enrichment_success_rate": 0.98
  }
}
```

## Troubleshooting

### Enrichment Not Working

```bash
# Check orchestrator connection
kubectl logs -f deployment/alert-processor-service | grep "orchestrator"

# Should see:
# "Connecting to orchestrator service" url=http://orchestrator-service:8000
# "Orchestrator query successful"

# If connection fails:
# 1. Verify orchestrator service is running
kubectl get pods -l app.kubernetes.io/name=orchestrator-service

# 2. Check service URL in config
kubectl get configmap bakery-config -o yaml | grep ORCHESTRATOR_SERVICE_URL

# 3. Test connection from pod
kubectl exec -it deployment/alert-processor-service -- curl http://orchestrator-service:8000/health
```

### Low Priority Scores

```bash
# Check priority calculation
kubectl logs -f deployment/alert-processor-service | grep "priority_score"

# Should see:
# "Alert enriched" priority_score=85 priority_level="important"

# If scores too low:
# 1. Check weights in config
kubectl get configmap bakery-config -o yaml | grep WEIGHT

# 2. Review business impact calculation
# - financial_impact_eur
# - affected_orders
# - urgency (time_until_deadline)
```

### Smart Actions Missing

```bash
# Check smart action generation
kubectl logs -f deployment/alert-processor-service | grep "smart_actions"

# Should see:
# "Generated smart actions" count=3 types=["phone","navigation","email"]

# If actions missing:
# 1. Check metadata in alert
# - supplier_id, supplier_phone (for phone actions)
# - ingredient_id, product_id (for navigation)
```

---

## Alert Escalation & Chaining System

### Overview

The Alert Processor implements sophisticated **time-based escalation** and **alert chaining** mechanisms to ensure critical issues don't get lost and that related alerts are properly grouped for user clarity.

**Key Features**:
- **Priority Escalation**: Automatically boost priority as alerts age
- **Deadline Proximity Boosting**: Increase urgency as deadlines approach
- **Alert Chaining**: Link related alerts (e.g., stock shortage → production delay)
- **Deduplication**: Prevent alert spam by merging similar alerts
- **Periodic Recalculation**: Cronjob refreshes priorities hourly

### Priority Escalation Rules

#### Time-Based Escalation

```python
# services/alert_processor/app/jobs/priority_recalculation.py

def calculate_escalation_boost(alert: Alert) -> int:
    """
    Calculate priority boost based on alert age and deadline proximity.

    Returns: Additional points to add to base priority_score (0-50)
    """
    boost = 0
    now = datetime.utcnow()

    # Age-based escalation
    if alert.alert_class == "action_needed":
        age_hours = (now - alert.action_created_at).total_seconds() / 3600

        if age_hours > 72:  # 3 days old
            boost += 20
        elif age_hours > 48:  # 2 days old
            boost += 10

    # Deadline proximity escalation
    if alert.deadline:
        hours_until_deadline = (alert.deadline - now).total_seconds() / 3600

        if hours_until_deadline <= 6:  # Critical: <6 hours
            boost += 30
        elif hours_until_deadline <= 24:  # Important: <24 hours
            boost += 15

    return min(boost, 50)  # Cap at +50 points
```

#### Escalation Cronjob

```yaml
# infrastructure/kubernetes/base/cronjobs/alert-priority-recalculation-cronjob.yaml
apiVersion: batch/v1
kind: CronJob
metadata:
  name: alert-priority-recalculation
spec:
  schedule: "15 * * * *"  # Hourly at :15
  concurrencyPolicy: Forbid
  jobTemplate:
    spec:
      activeDeadlineSeconds: 1800  # 30 min timeout
      template:
        spec:
          containers:
          - name: priority-recalculation
            image: alert-processor:latest
            command: ["python3", "-m", "app.jobs.priority_recalculation"]
            resources:
              requests:
                memory: "128Mi"
                cpu: "50m"
              limits:
                memory: "256Mi"
                cpu: "100m"
```

**Execution Flow**:
```
Hourly Trigger (minute :15)
  ↓
Query all alerts WHERE alert_class = 'action_needed' AND state != 'resolved'
  ↓
For each alert (batch size: 50):
  1. Calculate age in hours
  2. Calculate hours until deadline (if exists)
  3. Apply escalation rules
  4. Recalculate priority_score = base_score + escalation_boost
  5. Update priority_level (critical/important/info)
  6. Store escalation metadata
  7. UPDATE alerts SET priority_score, priority_level, escalation_metadata
  ↓
Invalidate Redis cache (tenant:{id}:alerts:priority)
  ↓
Next API call fetches updated priorities
```

**Performance**:
- Batch processing: 50 alerts at a time
- Typical execution: 100-500ms per tenant
- Multi-tenant scaling: ~1s per 100 active alerts

### Alert Chaining

#### Chain Types

**1. Causal Chains** - One alert causes another
```
LOW_STOCK_WARNING (ingredient: flour)
  ↓ (30 minutes later, inventory depletes)
CRITICAL_STOCK_SHORTAGE (ingredient: flour)
  ↓ (production cannot start)
PRODUCTION_DELAY (batch: baguettes, reason: missing flour)
  ↓ (order cannot fulfill)
ORDER_FULFILLMENT_RISK (order: #1234, missing: baguettes)
```

**2. Related Entity Chains** - Same entity, different aspects
```
PO_APPROVAL_NEEDED (po_id: 42)
  ↓ (approved, but delivery late)
DELIVERY_OVERDUE (po_id: 42)
  ↓ (delivery not received)
STOCK_RECEIPT_INCOMPLETE (po_id: 42)
```

**3. Temporal Chains** - Same issue over time
```
FORECAST_ACCURACY_LOW (product: croissant, MAPE: 35%)
  ↓ (7 days later, no improvement)
FORECAST_ACCURACY_LOW (product: croissant, MAPE: 34%)
  ↓ (14 days later, flagged for retraining)
MODEL_RETRAINING_NEEDED (product: croissant)
```

#### Chain Detection

```python
# services/alert_processor/app/services/enrichment/chaining.py

async def detect_alert_chain(alert: Alert) -> Optional[AlertChain]:
    """
    Detect if this alert is part of a chain.

    Returns: AlertChain object with parent/children or None
    """
    # Check for parent alert (what caused this?)
    parent = await find_parent_alert(alert)

    # Check for child alerts (what did this cause?)
    children = await find_child_alerts(alert)

    if parent or children:
        return AlertChain(
            chain_id=generate_chain_id(alert),
            root_alert_id=parent.id if parent else alert.id,
            parent_alert_id=parent.id if parent else None,
            child_alert_ids=[c.id for c in children],
            chain_type=classify_chain_type(alert, parent, children)
        )

    return None

async def find_parent_alert(alert: Alert) -> Optional[Alert]:
    """
    Find the alert that likely caused this one.

    Heuristics:
    - Same entity_id (e.g., ingredient, po_id)
    - Created within past 4 hours
    - Related event_type (e.g., LOW_STOCK → CRITICAL_STOCK_SHORTAGE)
    """
    if alert.event_type == "CRITICAL_STOCK_SHORTAGE":
        # Look for LOW_STOCK_WARNING on same ingredient
        return await db.query(Alert).filter(
            Alert.event_type == "LOW_STOCK_WARNING",
            Alert.entity_id == alert.entity_id,
            Alert.created_at >= alert.created_at - timedelta(hours=4),
            Alert.tenant_id == alert.tenant_id
        ).order_by(Alert.created_at.desc()).first()

    elif alert.event_type == "PRODUCTION_DELAY":
        # Look for CRITICAL_STOCK_SHORTAGE on ingredient in batch
        batch_ingredients = alert.context.get("missing_ingredients", [])
        return await find_stock_shortage_for_ingredients(batch_ingredients)

    # ... more chain detection rules

    return None
```

#### Chain Metadata

Chains are stored in alert metadata:
```json
{
  "chain": {
    "chain_id": "chain_flour_shortage_20251126",
    "chain_type": "causal",
    "position": "child",
    "root_alert_id": "alert_123",
    "parent_alert_id": "alert_123",
    "child_alert_ids": ["alert_789"],
    "depth": 2
  }
}
```

**Frontend Display**:
- UnifiedActionQueueCard shows chain icon when `alert.metadata.chain` exists
- Click chain icon → Expands to show full causal chain
- Root cause highlighted in bold
- Downstream impacts shown as tree

### Alert Deduplication

#### Deduplication Rules

```python
# services/alert_processor/app/services/enrichment/deduplication.py

async def check_duplicate(alert: Alert) -> Optional[Alert]:
    """
    Check if this alert is a duplicate of an existing one.

    Deduplication Keys:
    - event_type + entity_id + tenant_id
    - Time window: within past 24 hours
    - State: not resolved

    Returns: Existing alert if duplicate, None otherwise
    """
    existing = await db.query(Alert).filter(
        Alert.event_type == alert.event_type,
        Alert.entity_id == alert.entity_id,
        Alert.entity_type == alert.entity_type,
        Alert.tenant_id == alert.tenant_id,
        Alert.state != "resolved",
        Alert.created_at >= datetime.utcnow() - timedelta(hours=24)
    ).first()

    if existing:
        # Merge new data into existing alert
        await merge_duplicate(existing, alert)
        return existing

    return None

async def merge_duplicate(existing: Alert, new: Alert) -> None:
    """
    Merge duplicate alert into existing one.

    Updates:
    - occurrence_count += 1
    - last_occurrence_at = now
    - context = merge(existing.context, new.context)
    - priority_score = max(existing.priority_score, new.priority_score)
    """
    existing.occurrence_count += 1
    existing.last_occurrence_at = datetime.utcnow()
    existing.context = merge_contexts(existing.context, new.context)
    existing.priority_score = max(existing.priority_score, new.priority_score)

    # Add to metadata
    existing.metadata["duplicates"] = existing.metadata.get("duplicates", [])
    existing.metadata["duplicates"].append({
        "occurred_at": new.created_at.isoformat(),
        "context": new.context
    })

    await db.commit()
```

#### Deduplication Examples

**Example 1: Repeated Stock Warnings**
```
08:00 → LOW_STOCK_WARNING (flour, quantity: 50kg)
10:00 → LOW_STOCK_WARNING (flour, quantity: 45kg) → MERGED
12:00 → LOW_STOCK_WARNING (flour, quantity: 40kg) → MERGED

Result:
- Single alert with occurrence_count: 3
- Last_occurrence_at: 12:00
- Context shows quantity trend: 50kg → 45kg → 40kg
- Frontend displays: "Low stock warning (3 occurrences in 4 hours)"
```

**Example 2: Delivery Overdue Spam**
```
14:30 → DELIVERY_OVERDUE (PO-2025-043)
15:30 → DELIVERY_OVERDUE (PO-2025-043) → MERGED
16:30 → DELIVERY_OVERDUE (PO-2025-043) → MERGED

Result:
- Single critical alert
- Occurrence_count: 3
- Frontend: "Delivery overdue: PO-2025-043 (still pending after 2 hours)"
```

### Escalation Metadata

Each alert with escalation includes metadata:

```json
{
  "escalation": {
    "is_escalated": true,
    "escalation_level": 2,
    "escalation_reason": "Pending for 50 hours",
    "original_priority": 70,
    "current_priority": 80,
    "boost_applied": 10,
    "escalated_at": "2025-11-26T14:30:00Z",
    "escalation_history": [
      {
        "timestamp": "2025-11-24T12:00:00Z",
        "priority": 70,
        "boost": 0,
        "reason": "Initial priority"
      },
      {
        "timestamp": "2025-11-26T14:15:00Z",
        "priority": 80,
        "boost": 10,
        "reason": "Pending for 50 hours (>48h rule)"
      }
    ]
  }
}
```

**Frontend Display**:
- Badge showing "Escalated" with flame icon 🔥
- Tooltip: "Pending for 50 hours - Priority increased by 10 points"
- Timeline showing escalation history

### Integration with Priority Scoring

Escalation boost is **additive** to base priority:

```python
# Final priority calculation
base_priority = calculate_base_priority(alert)  # Multi-factor score (0-100)
escalation_boost = calculate_escalation_boost(alert)  # Age + deadline (0-50)
final_priority = min(base_priority + escalation_boost, 100)  # Capped at 100

# Update priority level
if final_priority >= 90:
    priority_level = "critical"
elif final_priority >= 70:
    priority_level = "important"
else:
    priority_level = "info"
```

**Example**:
```
Initial: DELIVERY_OVERDUE
- Base priority: 85 (critical event)
- Escalation boost: 0 (just created)
- Final priority: 85 (important)

After 2 hours (still unresolved):
- Base priority: 85 (unchanged)
- Escalation boost: 0 (not yet at 48h threshold)
- Final priority: 85 (important)

After 50 hours (still unresolved):
- Base priority: 85 (unchanged)
- Escalation boost: +10 (>48h rule)
- Final priority: 95 (critical)

After 50 hours + 23h to deadline:
- Base priority: 85 (unchanged)
- Escalation boost: +10 (age) + 15 (deadline <24h) = +25
- Final priority: 100 (critical, capped)
```

### Monitoring Escalations

**Metrics**:
- `alerts_escalated_total` - Count of alerts escalated
- `escalation_boost_avg` - Average boost applied
- `alerts_with_chains_total` - Count of chained alerts
- `deduplication_merge_total` - Count of merged duplicates

**Logs**:
```
[2025-11-26 14:15:03] INFO: Priority recalculation job started
[2025-11-26 14:15:04] INFO: Processing 45 action_needed alerts
[2025-11-26 14:15:05] INFO: Escalated alert alert_123: 70 → 80 (+10, pending 50h)
[2025-11-26 14:15:05] INFO: Escalated alert alert_456: 85 → 100 (+15, deadline in 20h)
[2025-11-26 14:15:06] INFO: Detected chain: alert_789 caused by alert_123
[2025-11-26 14:15:07] INFO: Merged duplicate: alert_999 into alert_998 (occurrence 3)
[2025-11-26 14:15:08] INFO: Priority recalculation completed in 5.2s
[2025-11-26 14:15:08] INFO: Invalidated Redis cache for 8 tenants
```

**Alerts** (for Ops team):
- Escalation rate >30% → Warning (too many stale alerts)
- Chain depth >5 → Warning (complex cascading failures)
- Deduplication rate >50% → Warning (alert spam)

### Testing

**Unit Tests**:
```python
# tests/jobs/test_priority_recalculation.py

async def test_escalation_48h_rule():
    # Given: Alert created 50 hours ago
    alert = create_test_alert(
        event_type="DELIVERY_OVERDUE",
        priority_score=85,
        action_created_at=now() - timedelta(hours=50)
    )

    # When: Recalculate priority
    boost = calculate_escalation_boost(alert)

    # Then: +10 boost applied
    assert boost == 10
    assert alert.priority_score == 95

async def test_chain_detection_stock_to_production():
    # Given: Low stock warning followed by production delay
    parent = create_test_alert(
        event_type="LOW_STOCK_WARNING",
        entity_id="ingredient_123"
    )
    child = create_test_alert(
        event_type="PRODUCTION_DELAY",
        context={"missing_ingredients": ["ingredient_123"]}
    )

    # When: Detect chain
    chain = await detect_alert_chain(child)

    # Then: Chain detected
    assert chain.parent_alert_id == parent.id
    assert chain.chain_type == "causal"

async def test_deduplication_merge():
    # Given: Existing alert
    existing = create_test_alert(
        event_type="LOW_STOCK_WARNING",
        entity_id="ingredient_123",
        occurrence_count=1
    )

    # When: Duplicate arrives
    duplicate = create_test_alert(
        event_type="LOW_STOCK_WARNING",
        entity_id="ingredient_123"
    )
    result = await check_duplicate(duplicate)

    # Then: Merged into existing
    assert result.id == existing.id
    assert result.occurrence_count == 2
```

### Performance Characteristics

**Priority Recalculation Cronjob**:
- Batch size: 50 alerts
- Execution time: 100-500ms per tenant
- Scaling: ~1s per 100 active alerts
- Multi-tenant (100 tenants, 1000 active alerts): ~10s

**Chain Detection**:
- Database queries: 2-3 per alert (parent + children lookup)
- Caching: Recent alerts cached in Redis
- Execution time: 50-150ms per alert

**Deduplication**:
- Database query: 1 per incoming alert
- Index on: (event_type, entity_id, tenant_id, created_at)
- Execution time: 10-30ms per check

### Configuration

**Environment Variables**:
```bash
# Escalation Rules
ESCALATION_AGE_48H_BOOST=10      # +10 points after 48 hours
ESCALATION_AGE_72H_BOOST=20      # +20 points after 72 hours
ESCALATION_DEADLINE_24H_BOOST=15 # +15 points if <24h to deadline
ESCALATION_DEADLINE_6H_BOOST=30  # +30 points if <6h to deadline
ESCALATION_MAX_BOOST=50          # Cap escalation boost

# Deduplication
DEDUPLICATION_WINDOW_HOURS=24    # Consider duplicates within 24h
DEDUPLICATION_MAX_OCCURRENCES=10 # Stop merging after 10 occurrences

# Chain Detection
CHAIN_DETECTION_WINDOW_HOURS=4   # Look for chains within 4h
CHAIN_MAX_DEPTH=5                # Warn if chain depth >5
```

---

**Copyright © 2025 Bakery-IA. All rights reserved.**