diff --git a/ANNOTATE/core/real_training_adapter.py b/ANNOTATE/core/real_training_adapter.py
index 8003a98..7bfa0a7 100644
--- a/ANNOTATE/core/real_training_adapter.py
+++ b/ANNOTATE/core/real_training_adapter.py
@@ -28,6 +28,73 @@ except ImportError:
 logger = logging.getLogger(__name__)
 
 
+def parse_timestamp_to_utc(timestamp_str: str) -> datetime:
+    """
+    Unified timestamp parser that handles all formats and ensures UTC timezone.
+    
+    Handles:
+    - ISO format with timezone: '2025-10-27T14:00:00+00:00'
+    - ISO format with Z: '2025-10-27T14:00:00Z'
+    - Space-separated with seconds: '2025-10-27 14:00:00'
+    - Space-separated without seconds: '2025-10-27 14:00'
+    
+    Args:
+        timestamp_str: Timestamp string in various formats
+        
+    Returns:
+        Timezone-aware datetime object in UTC
+        
+    Raises:
+        ValueError: If timestamp cannot be parsed
+    """
+    if not timestamp_str:
+        raise ValueError("Empty timestamp string")
+    
+    # Try ISO format first (handles T separator and timezone info)
+    if 'T' in timestamp_str or '+' in timestamp_str:
+        try:
+            # Handle 'Z' suffix (Zulu time = UTC)
+            if timestamp_str.endswith('Z'):
+                timestamp_str = timestamp_str[:-1] + '+00:00'
+            return datetime.fromisoformat(timestamp_str)
+        except ValueError:
+            pass
+    
+    # Try space-separated formats
+    # Replace space with T for fromisoformat compatibility
+    if ' ' in timestamp_str:
+        try:
+            # Try parsing with fromisoformat after converting space to T
+            dt = datetime.fromisoformat(timestamp_str.replace(' ', 'T'))
+            # Make timezone-aware if naive
+            if dt.tzinfo is None:
+                dt = dt.replace(tzinfo=timezone.utc)
+            return dt
+        except ValueError:
+            pass
+    
+    # Try explicit format parsing as fallback
+    formats = [
+        '%Y-%m-%d %H:%M:%S',  # With seconds
+        '%Y-%m-%d %H:%M',     # Without seconds
+        '%Y-%m-%dT%H:%M:%S',  # ISO without timezone
+        '%Y-%m-%dT%H:%M',     # ISO without seconds or timezone
+    ]
+    
+    for fmt in formats:
+        try:
+            dt = datetime.strptime(timestamp_str, fmt)
+            # Make timezone-aware
+            if dt.tzinfo is None:
+                dt = dt.replace(tzinfo=timezone.utc)
+            return dt
+        except ValueError:
+            continue
+    
+    # If all parsing attempts fail
+    raise ValueError(f"Could not parse timestamp: '{timestamp_str}'")
+
+
 @dataclass
 class TrainingSession:
     """Real training session tracking"""
@@ -214,7 +281,10 @@ class RealTrainingAdapter:
     
     def _fetch_market_state_for_test_case(self, test_case: Dict) -> Dict:
         """
-        Fetch market state dynamically for a test case
+        Fetch market state dynamically for a test case from DuckDB storage
+        
+        This fetches HISTORICAL data at the specific timestamp from the annotation,
+        not current/latest data.
         
         Args:
             test_case: Test case dictionary with timestamp, symbol, etc.
@@ -234,17 +304,32 @@ class RealTrainingAdapter:
                 logger.warning("No timestamp in test case")
                 return {}
             
-            # Parse timestamp
-            from datetime import datetime
-            timestamp = datetime.fromisoformat(timestamp_str.replace('Z', '+00:00'))
+            # Parse timestamp using unified parser
+            try:
+                timestamp = parse_timestamp_to_utc(timestamp_str)
+            except Exception as e:
+                logger.warning(f"Could not parse timestamp '{timestamp_str}': {e}")
+                return {}
             
             # Get training config
             training_config = test_case.get('training_config', {})
             timeframes = training_config.get('timeframes', ['1s', '1m', '1h', '1d'])
             context_window = training_config.get('context_window_minutes', 5)
+            negative_samples_window = training_config.get('negative_samples_window', 15)  # ±15 candles
             
-            logger.info(f"   Fetching market state for {symbol} at {timestamp}")
-            logger.info(f"   Timeframes: {timeframes}, Context window: {context_window} minutes")
+            # Calculate extended time range to include negative sampling window
+            # For 1m timeframe: ±15 candles = ±15 minutes
+            # Add buffer to ensure we have enough data
+            extended_window_minutes = max(context_window, negative_samples_window + 10)
+            
+            logger.info(f"   Fetching HISTORICAL market state for {symbol} at {timestamp}")
+            logger.info(f"   Timeframes: {timeframes}, Extended window: ±{extended_window_minutes} minutes")
+            logger.info(f"   (Includes ±{negative_samples_window} candles for negative sampling)")
+            
+            # Calculate time range for extended context window
+            from datetime import timedelta
+            start_time = timestamp - timedelta(minutes=extended_window_minutes)
+            end_time = timestamp + timedelta(minutes=extended_window_minutes)
             
             # Fetch data for each timeframe
             market_state = {
@@ -253,14 +338,67 @@ class RealTrainingAdapter:
                 'timeframes': {}
             }
             
+            # Try to get data from DuckDB storage first (historical data)
+            duckdb_storage = None
+            if hasattr(self.data_provider, 'duckdb_storage'):
+                duckdb_storage = self.data_provider.duckdb_storage
+            
             for timeframe in timeframes:
-                # Get historical data around the timestamp
-                # For now, just get the latest data (we can improve this later)
-                df = self.data_provider.get_historical_data(
-                    symbol=symbol,
-                    timeframe=timeframe,
-                    limit=100  # Get 100 candles for context
-                )
+                df = None
+                
+                # Calculate appropriate limit based on timeframe and window
+                # We want enough candles to cover the extended window plus negative samples
+                if timeframe == '1s':
+                    limit = extended_window_minutes * 60 * 2 + 100  # 2x for safety + buffer
+                elif timeframe == '1m':
+                    limit = extended_window_minutes * 2 + 50  # 2x for safety + buffer
+                elif timeframe == '1h':
+                    limit = max(200, extended_window_minutes // 30)  # At least 200 candles
+                elif timeframe == '1d':
+                    limit = 200  # Fixed for daily
+                else:
+                    limit = 300
+                
+                # Try DuckDB storage first (has historical data)
+                if duckdb_storage:
+                    try:
+                        df = duckdb_storage.get_ohlcv_data(
+                            symbol=symbol,
+                            timeframe=timeframe,
+                            start_time=start_time,
+                            end_time=end_time,
+                            limit=limit,
+                            direction='latest'
+                        )
+                        if df is not None and not df.empty:
+                            logger.debug(f"       {timeframe}: {len(df)} candles from DuckDB (historical)")
+                    except Exception as e:
+                        logger.debug(f"       {timeframe}: DuckDB query failed: {e}")
+                
+                # Fallback to data_provider (might have cached data)
+                if df is None or df.empty:
+                    try:
+                        # Use get_historical_data_replay for time-specific data
+                        replay_data = self.data_provider.get_historical_data_replay(
+                            symbol=symbol,
+                            start_time=start_time,
+                            end_time=end_time,
+                            timeframes=[timeframe]
+                        )
+                        df = replay_data.get(timeframe)
+                        if df is not None and not df.empty:
+                            logger.debug(f"       {timeframe}: {len(df)} candles from replay")
+                    except Exception as e:
+                        logger.debug(f"       {timeframe}: Replay failed: {e}")
+                
+                # Last resort: get latest data (not ideal but better than nothing)
+                if df is None or df.empty:
+                    logger.warning(f"       {timeframe}: No historical data found, using latest data as fallback")
+                    df = self.data_provider.get_historical_data(
+                        symbol=symbol,
+                        timeframe=timeframe,
+                        limit=limit  # Use calculated limit
+                    )
                 
                 if df is not None and not df.empty:
                     # Convert to dict format
@@ -272,15 +410,15 @@ class RealTrainingAdapter:
                         'close': df['close'].tolist(),
                         'volume': df['volume'].tolist()
                     }
-                    logger.debug(f"       {timeframe}: {len(df)} candles")
+                    logger.debug(f"       {timeframe}: {len(df)} candles stored")
                 else:
-                    logger.warning(f"       {timeframe}: No data")
+                    logger.warning(f"       {timeframe}: No data available")
             
             if market_state['timeframes']:
                 logger.info(f"    Fetched market state with {len(market_state['timeframes'])} timeframes")
                 return market_state
             else:
-                logger.warning(f"    No market data fetched")
+                logger.warning(f"    No market data fetched for any timeframe")
                 return {}
                 
         except Exception as e:
@@ -441,23 +579,9 @@ class RealTrainingAdapter:
                 logger.debug("      No holding period, skipping HOLD samples")
                 return hold_samples
             
-            # Parse entry timestamp - handle multiple formats
+            # Parse entry timestamp using unified parser
             try:
-                if 'T' in entry_timestamp:
-                    entry_time = datetime.fromisoformat(entry_timestamp.replace('Z', '+00:00'))
-                else:
-                    # Try with seconds first, then without
-                    try:
-                        entry_time = datetime.strptime(entry_timestamp, '%Y-%m-%d %H:%M:%S')
-                    except ValueError:
-                        # Try without seconds
-                        entry_time = datetime.strptime(entry_timestamp, '%Y-%m-%d %H:%M')
-                    
-                    # Make timezone-aware
-                    if pytz:
-                        entry_time = entry_time.replace(tzinfo=pytz.UTC)
-                    else:
-                        entry_time = entry_time.replace(tzinfo=timezone.utc)
+                entry_time = parse_timestamp_to_utc(entry_timestamp)
             except Exception as e:
                 logger.warning(f"Could not parse entry timestamp '{entry_timestamp}': {e}")
                 return hold_samples
@@ -473,18 +597,9 @@ class RealTrainingAdapter:
             
             # Find all candles between entry and exit
             for idx, ts_str in enumerate(timestamps):
-                # Parse timestamp and ensure it's timezone-aware
+                # Parse timestamp using unified parser
                 try:
-                    if 'T' in ts_str:
-                        ts = datetime.fromisoformat(ts_str.replace('Z', '+00:00'))
-                    else:
-                        ts = datetime.fromisoformat(ts_str.replace(' ', 'T'))
-                        # Make timezone-aware if it's naive
-                        if ts.tzinfo is None:
-                            if pytz:
-                                ts = ts.replace(tzinfo=pytz.UTC)
-                            else:
-                                ts = ts.replace(tzinfo=timezone.utc)
+                    ts = parse_timestamp_to_utc(ts_str)
                 except Exception as e:
                     logger.debug(f"Could not parse timestamp '{ts_str}': {e}")
                     continue
@@ -550,23 +665,9 @@ class RealTrainingAdapter:
             # Find the index of the signal timestamp
             from datetime import datetime
             
-            # Parse signal timestamp - handle different formats
+            # Parse signal timestamp using unified parser
             try:
-                if 'T' in signal_timestamp:
-                    signal_time = datetime.fromisoformat(signal_timestamp.replace('Z', '+00:00'))
-                else:
-                    # Try with seconds first, then without
-                    try:
-                        signal_time = datetime.strptime(signal_timestamp, '%Y-%m-%d %H:%M:%S')
-                    except ValueError:
-                        # Try without seconds
-                        signal_time = datetime.strptime(signal_timestamp, '%Y-%m-%d %H:%M')
-                    
-                    # Make timezone-aware
-                    if pytz:
-                        signal_time = signal_time.replace(tzinfo=pytz.UTC)
-                    else:
-                        signal_time = signal_time.replace(tzinfo=timezone.utc)
+                signal_time = parse_timestamp_to_utc(signal_timestamp)
             except Exception as e:
                 logger.warning(f"Could not parse signal timestamp '{signal_timestamp}': {e}")
                 return negative_samples
@@ -574,22 +675,8 @@ class RealTrainingAdapter:
             signal_index = None
             for idx, ts_str in enumerate(timestamps):
                 try:
-                    # Parse timestamp from market data - handle multiple formats
-                    if 'T' in ts_str:
-                        ts = datetime.fromisoformat(ts_str.replace('Z', '+00:00'))
-                    else:
-                        # Try with seconds first, then without
-                        try:
-                            ts = datetime.strptime(ts_str, '%Y-%m-%d %H:%M:%S')
-                        except ValueError:
-                            ts = datetime.strptime(ts_str, '%Y-%m-%d %H:%M')
-                        
-                        # Make timezone-aware if naive
-                        if ts.tzinfo is None:
-                            if pytz:
-                                ts = ts.replace(tzinfo=pytz.UTC)
-                            else:
-                                ts = ts.replace(tzinfo=timezone.utc)
+                    # Parse timestamp using unified parser
+                    ts = parse_timestamp_to_utc(ts_str)
                     
                     # Match within 1 minute
                     if abs((ts - signal_time).total_seconds()) < 60:
@@ -1147,8 +1234,9 @@ class RealTrainingAdapter:
             future_prices = torch.tensor([future_price], dtype=torch.float32)
             
             # Trade success (1.0 if profitable, 0.0 otherwise)
+            # Shape must be [batch_size, 1] to match confidence head output
             profit_loss_pct = training_sample.get('profit_loss_pct', 0.0)
-            trade_success = torch.tensor([1.0 if profit_loss_pct > 0 else 0.0], dtype=torch.float32)
+            trade_success = torch.tensor([[1.0 if profit_loss_pct > 0 else 0.0]], dtype=torch.float32)
             
             # Return batch dictionary
             batch = {
diff --git a/NN/models/advanced_transformer_trading.py b/NN/models/advanced_transformer_trading.py
index 4bdb97d..7832aa2 100644
--- a/NN/models/advanced_transformer_trading.py
+++ b/NN/models/advanced_transformer_trading.py
@@ -971,20 +971,20 @@ class TradingTransformerTrainer:
         
         # Add confidence loss if available
         if 'confidence' in outputs and 'trade_success' in batch:
-            # Ensure both tensors have compatible shapes for BCELoss
-            # BCELoss requires both inputs to have the same shape
-            confidence_pred = outputs['confidence']  # Keep as [batch_size, 1]
+            # Both tensors should have shape [batch_size, 1]
+            # confidence: [batch_size, 1] from confidence_head
+            # trade_success: [batch_size, 1] from batch preparation
+            confidence_pred = outputs['confidence']
             trade_target = batch['trade_success'].float()
             
-            # Reshape target to match prediction shape [batch_size, 1]
-            if trade_target.dim() == 1:
-                trade_target = trade_target.unsqueeze(-1)
-            
-            # Ensure both have same shape
+            # Verify shapes match (should both be [batch_size, 1])
             if confidence_pred.shape != trade_target.shape:
-                # If shapes still don't match, squeeze both to 1D
-                confidence_pred = confidence_pred.view(-1)
-                trade_target = trade_target.view(-1)
+                logger.warning(f"Shape mismatch: confidence {confidence_pred.shape} vs target {trade_target.shape}")
+                # Reshape to match if needed
+                if trade_target.dim() == 1:
+                    trade_target = trade_target.unsqueeze(-1)
+                if confidence_pred.dim() == 1:
+                    confidence_pred = confidence_pred.unsqueeze(-1)
             
             confidence_loss = self.confidence_criterion(confidence_pred, trade_target)
             # Use addition instead of += to avoid inplace operation
diff --git a/docs/UNIFIED_STORAGE_COMPLETE.md b/docs/UNIFIED_STORAGE_COMPLETE.md
deleted file mode 100644
index c0c2709..0000000
--- a/docs/UNIFIED_STORAGE_COMPLETE.md
+++ /dev/null
@@ -1,355 +0,0 @@
-# Unified Data Storage System - Complete Implementation
-
-## 🎉 Project Complete!
-
-The unified data storage system has been successfully implemented and integrated into the existing DataProvider.
-
-##  Completed Tasks (8 out of 10)
-
-### Task 1: TimescaleDB Schema and Infrastructure 
-**Files:**
-- `core/unified_storage_schema.py` - Schema manager with migrations
-- `scripts/setup_unified_storage.py` - Automated setup script
-- `docs/UNIFIED_STORAGE_SETUP.md` - Setup documentation
-
-**Features:**
-- 5 hypertables (OHLCV, order book, aggregations, imbalances, trades)
-- 5 continuous aggregates for multi-timeframe data
-- 15+ optimized indexes
-- Compression policies (>80% compression)
-- Retention policies (30 days to 2 years)
-
-### Task 2: Data Models and Validation 
-**Files:**
-- `core/unified_data_models.py` - Data structures
-- `core/unified_data_validator.py` - Validation logic
-
-**Features:**
-- `InferenceDataFrame` - Complete inference data
-- `OrderBookDataFrame` - Order book with imbalances
-- `OHLCVCandle`, `TradeEvent` - Individual data types
-- Comprehensive validation and sanitization
-
-### Task 3: Cache Layer 
-**Files:**
-- `core/unified_cache_manager.py` - In-memory caching
-
-**Features:**
-- <10ms read latency
-- 5-minute rolling window
-- Thread-safe operations
-- Automatic eviction
-- Statistics tracking
-
-### Task 4: Database Connection and Query Layer 
-**Files:**
-- `core/unified_database_manager.py` - Connection pool and queries
-
-**Features:**
-- Async connection pooling
-- Health monitoring
-- Optimized query methods
-- <100ms query latency
-- Multi-timeframe support
-
-### Task 5: Data Ingestion Pipeline 
-**Files:**
-- `core/unified_ingestion_pipeline.py` - Real-time ingestion
-
-**Features:**
-- Batch writes (100 items or 5 seconds)
-- Data validation before storage
-- Background flush worker
-- >1000 ops/sec throughput
-- Error handling and retry logic
-
-### Task 6: Unified Data Provider API 
-**Files:**
-- `core/unified_data_provider_extension.py` - Main API
-
-**Features:**
-- Single `get_inference_data()` endpoint
-- Automatic cache/database routing
-- Multi-timeframe data retrieval
-- Order book data access
-- Statistics tracking
-
-### Task 7: Data Migration System 
-**Status:** Skipped (decided to drop existing Parquet data)
-
-### Task 8: Integration with Existing DataProvider 
-**Files:**
-- `core/data_provider.py` - Updated with unified storage methods
-- `docs/UNIFIED_STORAGE_INTEGRATION.md` - Integration guide
-- `examples/unified_storage_example.py` - Usage examples
-
-**Features:**
-- Seamless integration with existing code
-- Backward compatible
-- Opt-in unified storage
-- Easy to enable/disable
-
-## 📊 System Architecture
-
-```
-┌─────────────────────────────────────────────┐
-│         Application Layer                    │
-│  (Models, Backtesting, Annotation, etc.)    │
-└────────────────┬────────────────────────────┘
-                 │
-                 ▼
-┌─────────────────────────────────────────────┐
-│         DataProvider (Existing)              │
-│  + Unified Storage Extension (New)          │
-└────────────────┬────────────────────────────┘
-                 │
-        ┌────────┴────────┐
-        ▼                 ▼
-┌──────────────┐   ┌──────────────┐
-│ Cache Layer  │   │ Database     │
-│ (In-Memory)  │   │ (TimescaleDB)│
-│              │   │              │
-│ - Last 5 min │   │ - Historical │
-│ - <10ms read │   │ - <100ms read│
-│ - Real-time  │   │ - Compressed │
-└──────────────┘   └──────────────┘
-```
-
-##  Key Features
-
-### Performance
--  Cache reads: <10ms
--  Database queries: <100ms
--  Ingestion: >1000 ops/sec
--  Compression: >80%
-
-### Reliability
--  Data validation
--  Error handling
--  Health monitoring
--  Statistics tracking
--  Automatic reconnection
-
-### Usability
--  Single endpoint for all data
--  Automatic routing (cache vs database)
--  Type-safe interfaces
--  Backward compatible
--  Easy to integrate
-
-## 📝 Quick Start
-
-### 1. Setup Database
-
-```bash
-python scripts/setup_unified_storage.py
-```
-
-### 2. Enable in Code
-
-```python
-from core.data_provider import DataProvider
-import asyncio
-
-data_provider = DataProvider()
-
-async def setup():
-    await data_provider.enable_unified_storage()
-
-asyncio.run(setup())
-```
-
-### 3. Use Unified API
-
-```python
-# Get real-time data (from cache)
-data = await data_provider.get_inference_data_unified('ETH/USDT')
-
-# Get historical data (from database)
-data = await data_provider.get_inference_data_unified(
-    'ETH/USDT',
-    timestamp=datetime(2024, 1, 15, 12, 30)
-)
-```
-
-## 📚 Documentation
-
-- **Setup Guide**: `docs/UNIFIED_STORAGE_SETUP.md`
-- **Integration Guide**: `docs/UNIFIED_STORAGE_INTEGRATION.md`
-- **Examples**: `examples/unified_storage_example.py`
-- **Design Document**: `.kiro/specs/unified-data-storage/design.md`
-- **Requirements**: `.kiro/specs/unified-data-storage/requirements.md`
-
-## 🎯 Use Cases
-
-### Real-Time Trading
-```python
-# Fast access to latest market data
-data = await data_provider.get_inference_data_unified('ETH/USDT')
-price = data.get_latest_price()
-```
-
-### Backtesting
-```python
-# Historical data at any timestamp
-data = await data_provider.get_inference_data_unified(
-    'ETH/USDT',
-    timestamp=target_time,
-    context_window_minutes=60
-)
-```
-
-### Data Annotation
-```python
-# Retrieve data at specific timestamps for labeling
-for timestamp in annotation_timestamps:
-    data = await data_provider.get_inference_data_unified(
-        'ETH/USDT',
-        timestamp=timestamp,
-        context_window_minutes=5
-    )
-    # Display and annotate
-```
-
-### Model Training
-```python
-# Get complete inference data for training
-data = await data_provider.get_inference_data_unified(
-    'ETH/USDT',
-    timestamp=training_timestamp
-)
-
-features = {
-    'ohlcv': data.ohlcv_1m.to_numpy(),
-    'indicators': data.indicators,
-    'imbalances': data.imbalances.to_numpy()
-}
-```
-
-## 📈 Performance Metrics
-
-### Cache Performance
-- Hit Rate: >90% (typical)
-- Read Latency: <10ms
-- Capacity: 5 minutes of data
-- Eviction: Automatic
-
-### Database Performance
-- Query Latency: <100ms (typical)
-- Write Throughput: >1000 ops/sec
-- Compression Ratio: >80%
-- Storage: Optimized with TimescaleDB
-
-### Ingestion Performance
-- Validation: All data validated
-- Batch Size: 100 items or 5 seconds
-- Error Rate: <0.1% (typical)
-- Retry: Automatic with backoff
-
-## 🔧 Configuration
-
-### Database Config (`config.yaml`)
-```yaml
-database:
-  host: localhost
-  port: 5432
-  name: trading_data
-  user: postgres
-  password: postgres
-  pool_size: 20
-```
-
-### Cache Config
-```python
-cache_manager = DataCacheManager(
-    cache_duration_seconds=300  # 5 minutes
-)
-```
-
-### Ingestion Config
-```python
-ingestion_pipeline = DataIngestionPipeline(
-    batch_size=100,
-    batch_timeout_seconds=5.0
-)
-```
-
-## 🎓 Examples
-
-Run the example script:
-```bash
-python examples/unified_storage_example.py
-```
-
-This demonstrates:
-1. Real-time data access
-2. Historical data retrieval
-3. Multi-timeframe queries
-4. Order book data
-5. Statistics tracking
-
-## 🔍 Monitoring
-
-### Get Statistics
-```python
-stats = data_provider.get_unified_storage_stats()
-
-print(f"Cache hit rate: {stats['cache']['hit_rate_percent']}%")
-print(f"DB queries: {stats['database']['total_queries']}")
-print(f"Ingestion rate: {stats['ingestion']['total_ingested']}")
-```
-
-### Check Health
-```python
-if data_provider.is_unified_storage_enabled():
-    print(" Unified storage is running")
-else:
-    print(" Unified storage is not enabled")
-```
-
-## 🚧 Remaining Tasks (Optional)
-
-### Task 9: Performance Optimization
-- Add detailed monitoring dashboards
-- Implement query caching
-- Optimize database indexes
-- Add performance alerts
-
-### Task 10: Documentation and Deployment
-- Create video tutorials
-- Add API reference documentation
-- Create deployment guides
-- Add monitoring setup
-
-## 🎉 Success Metrics
-
- **Completed**: 8 out of 10 major tasks (80%)  
- **Core Functionality**: 100% complete  
- **Integration**: Seamless with existing code  
- **Performance**: Meets all targets  
- **Documentation**: Comprehensive guides  
- **Examples**: Working code samples  
-
-## 🙏 Next Steps
-
-The unified storage system is **production-ready** and can be used immediately:
-
-1. **Setup Database**: Run `python scripts/setup_unified_storage.py`
-2. **Enable in Code**: Call `await data_provider.enable_unified_storage()`
-3. **Start Using**: Use `get_inference_data_unified()` for all data access
-4. **Monitor**: Check statistics with `get_unified_storage_stats()`
-
-## 📞 Support
-
-For issues or questions:
-1. Check documentation in `docs/`
-2. Review examples in `examples/`
-3. Check database setup: `python scripts/setup_unified_storage.py`
-4. Review logs for errors
-
----
-
-**Status**:  Production Ready  
-**Version**: 1.0.0  
-**Last Updated**: 2024  
-**Completion**: 80% (8/10 tasks)
diff --git a/docs/UNIFIED_STORAGE_INTEGRATION.md b/docs/UNIFIED_STORAGE_INTEGRATION.md
deleted file mode 100644
index 8da9df4..0000000
--- a/docs/UNIFIED_STORAGE_INTEGRATION.md
+++ /dev/null
@@ -1,398 +0,0 @@
-# Unified Storage System Integration Guide
-
-## Overview
-
-The unified storage system has been integrated into the existing `DataProvider` class, providing a single endpoint for both real-time and historical data access.
-
-## Key Features
-
- **Single Endpoint**: One method for all data access  
- **Automatic Routing**: Cache for real-time, database for historical  
- **Backward Compatible**: All existing methods still work  
- **Opt-In**: Only enabled when explicitly initialized  
- **Fast**: <10ms cache reads, <100ms database queries  
-
-## Quick Start
-
-### 1. Enable Unified Storage
-
-```python
-from core.data_provider import DataProvider
-import asyncio
-
-# Create DataProvider (existing code works as before)
-data_provider = DataProvider()
-
-# Enable unified storage system
-async def setup():
-    success = await data_provider.enable_unified_storage()
-    if success:
-        print(" Unified storage enabled!")
-    else:
-        print(" Failed to enable unified storage")
-
-asyncio.run(setup())
-```
-
-### 2. Get Real-Time Data (from cache)
-
-```python
-async def get_realtime_data():
-    # Get latest real-time data (timestamp=None)
-    inference_data = await data_provider.get_inference_data_unified('ETH/USDT')
-    
-    print(f"Symbol: {inference_data.symbol}")
-    print(f"Timestamp: {inference_data.timestamp}")
-    print(f"Latest price: {inference_data.get_latest_price()}")
-    print(f"Data source: {inference_data.data_source}")  # 'cache'
-    print(f"Query latency: {inference_data.query_latency_ms}ms")  # <10ms
-    
-    # Check data completeness
-    if inference_data.has_complete_data():
-        print("✓ All required data present")
-    
-    # Get data summary
-    summary = inference_data.get_data_summary()
-    print(f"OHLCV 1m rows: {summary['ohlcv_1m_rows']}")
-    print(f"Has orderbook: {summary['has_orderbook']}")
-    print(f"Imbalances rows: {summary['imbalances_rows']}")
-
-asyncio.run(get_realtime_data())
-```
-
-### 3. Get Historical Data (from database)
-
-```python
-from datetime import datetime, timedelta
-
-async def get_historical_data():
-    # Get historical data at specific timestamp
-    target_time = datetime.now() - timedelta(hours=1)
-    
-    inference_data = await data_provider.get_inference_data_unified(
-        symbol='ETH/USDT',
-        timestamp=target_time,
-        context_window_minutes=5  # ±5 minutes of context
-    )
-    
-    print(f"Data source: {inference_data.data_source}")  # 'database'
-    print(f"Query latency: {inference_data.query_latency_ms}ms")  # <100ms
-    
-    # Access multi-timeframe data
-    print(f"1s candles: {len(inference_data.ohlcv_1s)}")
-    print(f"1m candles: {len(inference_data.ohlcv_1m)}")
-    print(f"1h candles: {len(inference_data.ohlcv_1h)}")
-    
-    # Access technical indicators
-    print(f"RSI: {inference_data.indicators.get('rsi_14')}")
-    print(f"MACD: {inference_data.indicators.get('macd')}")
-    
-    # Access context data
-    if inference_data.context_data is not None:
-        print(f"Context data: {len(inference_data.context_data)} rows")
-
-asyncio.run(get_historical_data())
-```
-
-### 4. Get Multi-Timeframe Data
-
-```python
-async def get_multi_timeframe():
-    # Get multiple timeframes at once
-    multi_tf = await data_provider.get_multi_timeframe_data_unified(
-        symbol='ETH/USDT',
-        timeframes=['1m', '5m', '1h'],
-        limit=100
-    )
-    
-    for timeframe, df in multi_tf.items():
-        print(f"{timeframe}: {len(df)} candles")
-        if not df.empty:
-            print(f"  Latest close: {df.iloc[-1]['close_price']}")
-
-asyncio.run(get_multi_timeframe())
-```
-
-### 5. Get Order Book Data
-
-```python
-async def get_orderbook():
-    # Get order book with imbalances
-    orderbook = await data_provider.get_order_book_data_unified('ETH/USDT')
-    
-    print(f"Mid price: {orderbook.mid_price}")
-    print(f"Spread: {orderbook.spread}")
-    print(f"Spread (bps): {orderbook.get_spread_bps()}")
-    
-    # Get best bid/ask
-    best_bid = orderbook.get_best_bid()
-    best_ask = orderbook.get_best_ask()
-    print(f"Best bid: {best_bid}")
-    print(f"Best ask: {best_ask}")
-    
-    # Get imbalance summary
-    imbalances = orderbook.get_imbalance_summary()
-    print(f"Imbalances: {imbalances}")
-
-asyncio.run(get_orderbook())
-```
-
-### 6. Get Statistics
-
-```python
-# Get unified storage statistics
-stats = data_provider.get_unified_storage_stats()
-
-print("=== Cache Statistics ===")
-print(f"Hit rate: {stats['cache']['hit_rate_percent']}%")
-print(f"Total entries: {stats['cache']['total_entries']}")
-
-print("\n=== Database Statistics ===")
-print(f"Total queries: {stats['database']['total_queries']}")
-print(f"Avg query time: {stats['database']['avg_query_time_ms']}ms")
-
-print("\n=== Ingestion Statistics ===")
-print(f"Total ingested: {stats['ingestion']['total_ingested']}")
-print(f"Validation failures: {stats['ingestion']['validation_failures']}")
-```
-
-## Integration with Existing Code
-
-### Backward Compatibility
-
-All existing DataProvider methods continue to work:
-
-```python
-# Existing methods still work
-df = data_provider.get_historical_data('ETH/USDT', '1m', limit=100)
-price = data_provider.get_current_price('ETH/USDT')
-features = data_provider.get_feature_matrix('ETH/USDT')
-
-# New unified methods available alongside
-inference_data = await data_provider.get_inference_data_unified('ETH/USDT')
-```
-
-### Gradual Migration
-
-You can migrate to unified storage gradually:
-
-```python
-# Option 1: Use existing methods (no changes needed)
-df = data_provider.get_historical_data('ETH/USDT', '1m')
-
-# Option 2: Use unified storage for new features
-inference_data = await data_provider.get_inference_data_unified('ETH/USDT')
-```
-
-## Use Cases
-
-### 1. Real-Time Trading
-
-```python
-async def realtime_trading_loop():
-    while True:
-        # Get latest market data (fast!)
-        data = await data_provider.get_inference_data_unified('ETH/USDT')
-        
-        # Make trading decision
-        if data.has_complete_data():
-            price = data.get_latest_price()
-            rsi = data.indicators.get('rsi_14', 50)
-            
-            if rsi < 30:
-                print(f"Buy signal at {price}")
-            elif rsi > 70:
-                print(f"Sell signal at {price}")
-        
-        await asyncio.sleep(1)
-```
-
-### 2. Backtesting
-
-```python
-async def backtest_strategy(start_time, end_time):
-    current_time = start_time
-    
-    while current_time < end_time:
-        # Get historical data at specific time
-        data = await data_provider.get_inference_data_unified(
-            'ETH/USDT',
-            timestamp=current_time,
-            context_window_minutes=60
-        )
-        
-        # Run strategy
-        if data.has_complete_data():
-            # Your strategy logic here
-            pass
-        
-        # Move to next timestamp
-        current_time += timedelta(minutes=1)
-```
-
-### 3. Data Annotation
-
-```python
-async def annotate_data(timestamps):
-    annotations = []
-    
-    for timestamp in timestamps:
-        # Get data at specific timestamp
-        data = await data_provider.get_inference_data_unified(
-            'ETH/USDT',
-            timestamp=timestamp,
-            context_window_minutes=5
-        )
-        
-        # Display to user for annotation
-        # User marks buy/sell signals
-        annotation = {
-            'timestamp': timestamp,
-            'price': data.get_latest_price(),
-            'signal': 'buy',  # User input
-            'data': data.to_dict()
-        }
-        annotations.append(annotation)
-    
-    return annotations
-```
-
-### 4. Model Training
-
-```python
-async def prepare_training_data(symbol, start_time, end_time):
-    training_samples = []
-    
-    current_time = start_time
-    while current_time < end_time:
-        # Get complete inference data
-        data = await data_provider.get_inference_data_unified(
-            symbol,
-            timestamp=current_time,
-            context_window_minutes=10
-        )
-        
-        if data.has_complete_data():
-            # Extract features
-            features = {
-                'ohlcv_1m': data.ohlcv_1m.to_numpy(),
-                'indicators': data.indicators,
-                'imbalances': data.imbalances.to_numpy(),
-                'orderbook': data.orderbook_snapshot
-            }
-            
-            training_samples.append(features)
-        
-        current_time += timedelta(minutes=1)
-    
-    return training_samples
-```
-
-## Configuration
-
-### Database Configuration
-
-Update `config.yaml`:
-
-```yaml
-database:
-  host: localhost
-  port: 5432
-  name: trading_data
-  user: postgres
-  password: postgres
-  pool_size: 20
-```
-
-### Setup Database
-
-```bash
-# Run setup script
-python scripts/setup_unified_storage.py
-```
-
-## Performance Tips
-
-1. **Use Real-Time Endpoint for Latest Data**
-   ```python
-   # Fast (cache)
-   data = await data_provider.get_inference_data_unified('ETH/USDT')
-   
-   # Slower (database)
-   data = await data_provider.get_inference_data_unified('ETH/USDT', datetime.now())
-   ```
-
-2. **Batch Historical Queries**
-   ```python
-   # Get multiple timeframes at once
-   multi_tf = await data_provider.get_multi_timeframe_data_unified(
-       'ETH/USDT',
-       ['1m', '5m', '1h'],
-       limit=100
-   )
-   ```
-
-3. **Monitor Performance**
-   ```python
-   stats = data_provider.get_unified_storage_stats()
-   print(f"Cache hit rate: {stats['cache']['hit_rate_percent']}%")
-   print(f"Avg query time: {stats['database']['avg_query_time_ms']}ms")
-   ```
-
-## Troubleshooting
-
-### Unified Storage Not Available
-
-```python
-if not data_provider.is_unified_storage_enabled():
-    success = await data_provider.enable_unified_storage()
-    if not success:
-        print("Check database connection and configuration")
-```
-
-### Slow Queries
-
-```python
-# Check query latency
-data = await data_provider.get_inference_data_unified('ETH/USDT', timestamp)
-if data.query_latency_ms > 100:
-    print(f"Slow query: {data.query_latency_ms}ms")
-    # Check database stats
-    stats = data_provider.get_unified_storage_stats()
-    print(stats['database'])
-```
-
-### Missing Data
-
-```python
-data = await data_provider.get_inference_data_unified('ETH/USDT', timestamp)
-if not data.has_complete_data():
-    summary = data.get_data_summary()
-    print(f"Missing data: {summary}")
-```
-
-## API Reference
-
-### Main Methods
-
-- `enable_unified_storage()` - Enable unified storage system
-- `disable_unified_storage()` - Disable unified storage system
-- `get_inference_data_unified()` - Get complete inference data
-- `get_multi_timeframe_data_unified()` - Get multi-timeframe data
-- `get_order_book_data_unified()` - Get order book with imbalances
-- `get_unified_storage_stats()` - Get statistics
-- `is_unified_storage_enabled()` - Check if enabled
-
-### Data Models
-
-- `InferenceDataFrame` - Complete inference data structure
-- `OrderBookDataFrame` - Order book with imbalances
-- `OHLCVCandle` - Single candlestick
-- `TradeEvent` - Individual trade
-
-## Support
-
-For issues or questions:
-1. Check database connection: `python scripts/setup_unified_storage.py`
-2. Review logs for errors
-3. Check statistics: `data_provider.get_unified_storage_stats()`
diff --git a/docs/UNIFIED_STORAGE_SETUP.md b/docs/UNIFIED_STORAGE_SETUP.md
deleted file mode 100644
index c85293f..0000000
--- a/docs/UNIFIED_STORAGE_SETUP.md
+++ /dev/null
@@ -1,337 +0,0 @@
-# Unified Data Storage Setup Guide
-
-## Overview
-
-The unified data storage system consolidates all market data storage into a single TimescaleDB backend, replacing fragmented Parquet files, pickle files, and in-memory caches.
-
-## Prerequisites
-
-### 1. PostgreSQL with TimescaleDB
-
-You need PostgreSQL 12+ with TimescaleDB extension installed.
-
-#### Installation Options
-
-**Option A: Docker (Recommended)**
-```bash
-docker run -d --name timescaledb \
-  -p 5432:5432 \
-  -e POSTGRES_PASSWORD=postgres \
-  -e POSTGRES_DB=trading_data \
-  timescale/timescaledb:latest-pg14
-```
-
-**Option B: Local Installation**
-- Follow TimescaleDB installation guide: https://docs.timescale.com/install/latest/
-- Create database: `createdb trading_data`
-
-### 2. Python Dependencies
-
-Ensure you have the required Python packages:
-```bash
-pip install asyncpg
-```
-
-## Database Configuration
-
-Update your `config.yaml` with database connection details:
-
-```yaml
-database:
-  host: localhost
-  port: 5432
-  name: trading_data
-  user: postgres
-  password: postgres
-  pool_size: 20
-```
-
-## Setup Process
-
-### Step 1: Run Setup Script
-
-```bash
-python scripts/setup_unified_storage.py
-```
-
-This script will:
-1. Connect to the database
-2. Verify TimescaleDB extension
-3. Create all required tables
-4. Convert tables to hypertables
-5. Create indexes for performance
-6. Set up continuous aggregates
-7. Configure compression policies
-8. Configure retention policies
-9. Verify the setup
-10. Run basic operation tests
-
-### Step 2: Verify Setup
-
-The setup script will display schema information:
-
-```
-=== Schema Information ===
-Migrations applied: 8
-Tables created: 5
-Hypertables: 5
-Continuous aggregates: 5
-
-=== Table Sizes ===
-  ohlcv_data: 8192 bytes
-  order_book_snapshots: 8192 bytes
-  order_book_1s_agg: 8192 bytes
-  order_book_imbalances: 8192 bytes
-  trade_events: 8192 bytes
-
-=== Hypertables ===
-  ohlcv_data: 0 chunks, compression=enabled
-  order_book_snapshots: 0 chunks, compression=enabled
-  order_book_1s_agg: 0 chunks, compression=enabled
-  order_book_imbalances: 0 chunks, compression=enabled
-  trade_events: 0 chunks, compression=enabled
-
-=== Continuous Aggregates ===
-  ohlcv_1m_continuous: 8192 bytes
-  ohlcv_5m_continuous: 8192 bytes
-  ohlcv_15m_continuous: 8192 bytes
-  ohlcv_1h_continuous: 8192 bytes
-  ohlcv_1d_continuous: 8192 bytes
-```
-
-## Database Schema
-
-### Tables
-
-#### 1. ohlcv_data
-Stores candlestick data for all timeframes with pre-calculated technical indicators.
-
-**Columns:**
-- `timestamp` (TIMESTAMPTZ): Candle timestamp
-- `symbol` (VARCHAR): Trading pair (e.g., 'ETH/USDT')
-- `timeframe` (VARCHAR): Timeframe (1s, 1m, 5m, 15m, 1h, 1d)
-- `open_price`, `high_price`, `low_price`, `close_price` (DECIMAL): OHLC prices
-- `volume` (DECIMAL): Trading volume
-- `trade_count` (INTEGER): Number of trades
-- Technical indicators: `rsi_14`, `macd`, `macd_signal`, `bb_upper`, `bb_middle`, `bb_lower`, etc.
-
-**Primary Key:** `(timestamp, symbol, timeframe)`
-
-#### 2. order_book_snapshots
-Stores raw order book snapshots.
-
-**Columns:**
-- `timestamp` (TIMESTAMPTZ): Snapshot timestamp
-- `symbol` (VARCHAR): Trading pair
-- `exchange` (VARCHAR): Exchange name
-- `bids` (JSONB): Bid levels (top 50)
-- `asks` (JSONB): Ask levels (top 50)
-- `mid_price`, `spread`, `bid_volume`, `ask_volume` (DECIMAL): Calculated metrics
-
-**Primary Key:** `(timestamp, symbol, exchange)`
-
-#### 3. order_book_1s_agg
-Stores 1-second aggregated order book data with $1 price buckets.
-
-**Columns:**
-- `timestamp` (TIMESTAMPTZ): Aggregation timestamp
-- `symbol` (VARCHAR): Trading pair
-- `price_bucket` (DECIMAL): Price bucket ($1 increments)
-- `bid_volume`, `ask_volume` (DECIMAL): Aggregated volumes
-- `bid_count`, `ask_count` (INTEGER): Number of orders
-- `imbalance` (DECIMAL): Order book imbalance
-
-**Primary Key:** `(timestamp, symbol, price_bucket)`
-
-#### 4. order_book_imbalances
-Stores multi-timeframe order book imbalance metrics.
-
-**Columns:**
-- `timestamp` (TIMESTAMPTZ): Calculation timestamp
-- `symbol` (VARCHAR): Trading pair
-- `imbalance_1s`, `imbalance_5s`, `imbalance_15s`, `imbalance_60s` (DECIMAL): Imbalances
-- `volume_imbalance_1s`, `volume_imbalance_5s`, etc. (DECIMAL): Volume-weighted imbalances
-- `price_range` (DECIMAL): Price range used for calculation
-
-**Primary Key:** `(timestamp, symbol)`
-
-#### 5. trade_events
-Stores individual trade events.
-
-**Columns:**
-- `timestamp` (TIMESTAMPTZ): Trade timestamp
-- `symbol` (VARCHAR): Trading pair
-- `exchange` (VARCHAR): Exchange name
-- `price` (DECIMAL): Trade price
-- `size` (DECIMAL): Trade size
-- `side` (VARCHAR): Trade side ('buy' or 'sell')
-- `trade_id` (VARCHAR): Unique trade identifier
-
-**Primary Key:** `(timestamp, symbol, exchange, trade_id)`
-
-### Continuous Aggregates
-
-Continuous aggregates automatically generate higher timeframe data from lower timeframes:
-
-1. **ohlcv_1m_continuous**: 1-minute candles from 1-second data
-2. **ohlcv_5m_continuous**: 5-minute candles from 1-minute data
-3. **ohlcv_15m_continuous**: 15-minute candles from 5-minute data
-4. **ohlcv_1h_continuous**: 1-hour candles from 15-minute data
-5. **ohlcv_1d_continuous**: 1-day candles from 1-hour data
-
-### Compression Policies
-
-Data is automatically compressed to save storage:
-
-- **ohlcv_data**: Compress after 7 days
-- **order_book_snapshots**: Compress after 1 day
-- **order_book_1s_agg**: Compress after 2 days
-- **order_book_imbalances**: Compress after 2 days
-- **trade_events**: Compress after 7 days
-
-Expected compression ratio: **>80%**
-
-### Retention Policies
-
-Old data is automatically deleted:
-
-- **ohlcv_data**: Retain for 2 years
-- **order_book_snapshots**: Retain for 30 days
-- **order_book_1s_agg**: Retain for 60 days
-- **order_book_imbalances**: Retain for 60 days
-- **trade_events**: Retain for 90 days
-
-## Performance Optimization
-
-### Indexes
-
-All tables have optimized indexes for common query patterns:
-
-- Symbol + timestamp queries
-- Timeframe-specific queries
-- Exchange-specific queries
-- Multi-column composite indexes
-
-### Query Performance Targets
-
-- **Cache reads**: <10ms
-- **Single timestamp queries**: <100ms
-- **Time range queries (1 hour)**: <500ms
-- **Ingestion throughput**: >1000 ops/sec
-
-### Best Practices
-
-1. **Use time_bucket for aggregations**:
-   ```sql
-   SELECT time_bucket('1 minute', timestamp) AS bucket,
-          symbol,
-          avg(close_price) AS avg_price
-   FROM ohlcv_data
-   WHERE symbol = 'ETH/USDT'
-     AND timestamp >= NOW() - INTERVAL '1 hour'
-   GROUP BY bucket, symbol;
-   ```
-
-2. **Query specific timeframes**:
-   ```sql
-   SELECT * FROM ohlcv_data
-   WHERE symbol = 'ETH/USDT'
-     AND timeframe = '1m'
-     AND timestamp >= NOW() - INTERVAL '1 day'
-   ORDER BY timestamp DESC;
-   ```
-
-3. **Use continuous aggregates for historical data**:
-   ```sql
-   SELECT * FROM ohlcv_1h_continuous
-   WHERE symbol = 'ETH/USDT'
-     AND timestamp >= NOW() - INTERVAL '7 days'
-   ORDER BY timestamp DESC;
-   ```
-
-## Monitoring
-
-### Check Database Size
-
-```sql
-SELECT 
-    hypertable_name,
-    pg_size_pretty(total_bytes) AS total_size,
-    pg_size_pretty(compressed_total_bytes) AS compressed_size,
-    ROUND((1 - compressed_total_bytes::numeric / total_bytes::numeric) * 100, 2) AS compression_ratio
-FROM timescaledb_information.hypertables
-WHERE hypertable_schema = 'public';
-```
-
-### Check Chunk Information
-
-```sql
-SELECT 
-    hypertable_name,
-    num_chunks,
-    num_compressed_chunks,
-    compression_enabled
-FROM timescaledb_information.hypertables
-WHERE hypertable_schema = 'public';
-```
-
-### Check Continuous Aggregate Status
-
-```sql
-SELECT 
-    view_name,
-    materialization_hypertable_name,
-    pg_size_pretty(total_bytes) AS size
-FROM timescaledb_information.continuous_aggregates
-WHERE view_schema = 'public';
-```
-
-## Troubleshooting
-
-### TimescaleDB Extension Not Found
-
-If you see "TimescaleDB extension not found":
-
-1. Ensure TimescaleDB is installed
-2. Connect to database and run: `CREATE EXTENSION timescaledb;`
-3. Restart the setup script
-
-### Connection Refused
-
-If you see "connection refused":
-
-1. Check PostgreSQL is running: `pg_isready`
-2. Verify connection details in `config.yaml`
-3. Check firewall settings
-
-### Permission Denied
-
-If you see "permission denied":
-
-1. Ensure database user has CREATE privileges
-2. Grant privileges: `GRANT ALL PRIVILEGES ON DATABASE trading_data TO postgres;`
-
-### Slow Queries
-
-If queries are slow:
-
-1. Check if indexes exist: `\di` in psql
-2. Analyze query plan: `EXPLAIN ANALYZE <your query>`
-3. Ensure compression is enabled
-4. Consider adding more specific indexes
-
-## Next Steps
-
-After setup is complete:
-
-1. **Implement data models** (Task 2)
-2. **Implement cache layer** (Task 3)
-3. **Implement database connection layer** (Task 4)
-4. **Start data migration** from Parquet files (Task 7)
-
-## Support
-
-For issues or questions:
-- Check TimescaleDB docs: https://docs.timescale.com/
-- Review PostgreSQL logs: `tail -f /var/log/postgresql/postgresql-*.log`
-- Enable debug logging in setup script
diff --git a/docs/main/MODEL_INPUTS_OUTPUTS_REFERENCE.md b/docs/main/MODEL_INPUTS_OUTPUTS_REFERENCE.md
new file mode 100644
index 0000000..44e5cf3
--- /dev/null
+++ b/docs/main/MODEL_INPUTS_OUTPUTS_REFERENCE.md
@@ -0,0 +1,308 @@
+# Model Inputs & Outputs Reference
+
+Quick reference for all trading models in the system.
+
+---
+
+## 1. Transformer (AdvancedTradingTransformer)
+
+**Type**: Sequence-to-sequence transformer for multi-timeframe analysis  
+**Size**: 46M parameters  
+**Architecture**: 12 layers, 16 attention heads, 1024 model dimension
+
+### Inputs
+```python
+price_data:   [batch, 150, 5]      # OHLCV sequences (150 candles)
+cob_data:     [batch, 150, 100]    # Change of Bid features
+tech_data:    [batch, 40]          # Technical indicators (SMA, returns, volatility)
+market_data:  [batch, 30]          # Market context (volume, pivots, support/resistance)
+```
+
+### Outputs
+```python
+action_logits:        [batch, 3]         # Raw logits for BUY(1), SELL(2), HOLD(0)
+action_probs:         [batch, 3]         # Softmax probabilities
+confidence:           [batch, 1]         # Trade confidence (0-1)
+price_prediction:     [batch, 1]         # Future price target
+volatility_prediction:[batch, 1]         # Expected volatility
+trend_strength:       [batch, 1]         # Trend strength (-1 to 1)
+
+# Next candle predictions for each timeframe
+next_candles: {
+    '1s': [batch, 5],  # [open, high, low, close, volume]
+    '1m': [batch, 5],
+    '1h': [batch, 5],
+    '1d': [batch, 5]
+}
+
+# Pivot point predictions (L1-L5)
+next_pivots: {
+    'L1': {
+        'price': [batch, 1],
+        'type_prob_high': [batch, 1],  # Probability of high pivot
+        'type_prob_low': [batch, 1],   # Probability of low pivot
+        'confidence': [batch, 1]
+    },
+    # ... L2, L3, L4, L5 (same structure)
+}
+
+# Trend vector analysis
+trend_analysis: {
+    'angle_radians': [batch, 1],    # Trend angle
+    'steepness': [batch, 1],        # Trend steepness
+    'direction': [batch, 1]         # Direction (-1 to 1)
+}
+```
+
+### Training Targets
+```python
+actions:        [batch]      # Action labels (0=HOLD, 1=BUY, 2=SELL)
+future_prices:  [batch]      # Price targets
+trade_success:  [batch, 1]   # Success labels (0.0 or 1.0)
+```
+
+---
+
+## 2. CNN (StandardizedCNN / EnhancedCNN)
+
+**Type**: Convolutional neural network for pattern recognition  
+**Size**: ~5-10M parameters  
+**Architecture**: Multi-scale convolutions with attention
+
+### Inputs
+```python
+# Via BaseDataInput.get_feature_vector()
+feature_vector: [batch, 7834]  # Flattened features containing:
+    - OHLCV ETH: 300 frames × 4 timeframes × 5 = 6000
+    - OHLCV BTC: 300 frames × 5 = 1500
+    - COB features: 184 (±20 buckets + MA imbalance)
+    - Technical indicators: 100 (padded)
+    - Last predictions: 50 (padded)
+```
+
+### Outputs
+```python
+action_logits:  [batch, 3]      # BUY, SELL, HOLD logits
+action_probs:   [batch, 3]      # Softmax probabilities
+confidence:     [batch, 1]      # Prediction confidence
+hidden_states:  [batch, 1024]   # Feature embeddings (for cross-model feeding)
+predicted_returns: [batch, 4]   # [return_1s, return_1m, return_1h, return_1d]
+```
+
+### Training Targets
+```python
+actions:    [batch]      # Action labels (0=HOLD, 1=BUY, 2=SELL)
+returns:    [batch, 4]   # Actual returns per timeframe
+```
+
+---
+
+## 3. DQN (Deep Q-Network Agent)
+
+**Type**: Reinforcement learning agent for sequential decision making  
+**Size**: ~15M parameters  
+**Architecture**: Deep Q-Network with dueling architecture
+
+### Inputs
+```python
+# Via BaseDataInput.get_feature_vector()
+state: [batch, 7850]  # Full feature vector including:
+    - Multi-timeframe OHLCV data
+    - COB features
+    - Technical indicators
+    - Market regime indicators
+    - Previous predictions
+```
+
+### Outputs
+```python
+q_values:       [batch, 3]      # Q-values for BUY, SELL, HOLD
+action:         int             # Selected action (0, 1, 2)
+confidence:     float           # Action confidence (0-1)
+
+# Auxiliary outputs
+regime_probs:   [batch, 4]      # [trending, ranging, volatile, mixed]
+price_direction:[batch, 3]      # [down, neutral, up]
+volatility:     [batch, 1]      # Predicted volatility
+value:          [batch, 1]      # State value (V)
+advantage:      [batch, 3]      # Action advantages (A)
+```
+
+### Training Targets
+```python
+# RL uses experience replay
+experience: {
+    'state': [7850],
+    'action': int,
+    'reward': float,
+    'next_state': [7850],
+    'done': bool
+}
+```
+
+---
+
+## 4. COB RL Model (MassiveRLNetwork)
+
+**Type**: Specialized RL for Change of Bid (COB) data  
+**Size**: ~3M parameters  
+**Architecture**: Deep network focused on order book dynamics
+
+### Inputs
+```python
+cob_features: [batch, input_size]  # COB-specific features:
+    - Bid/ask imbalance
+    - Order book depth
+    - Price level changes
+    - Volume at price levels
+    - Moving averages of imbalance
+```
+
+### Outputs
+```python
+price_logits:   [batch, 3]      # Direction logits [DOWN, SIDEWAYS, UP]
+price_probs:    [batch, 3]      # Direction probabilities
+confidence:     [batch, 1]      # Prediction confidence
+value:          [batch, 1]      # State value estimate
+predicted_direction: int        # 0=DOWN, 1=SIDEWAYS, 2=UP
+```
+
+### Training Targets
+```python
+targets: {
+    'direction': [batch],       # Direction labels (0, 1, 2)
+    'value': [batch],          # Value targets
+    'confidence': [batch]      # Confidence targets
+}
+```
+
+---
+
+## 5. Extrema Trainer
+
+**Type**: Pivot point detection and prediction  
+**Size**: ~1M parameters (lightweight)  
+**Architecture**: Statistical + ML hybrid
+
+### Inputs
+```python
+# Context data (200 candles)
+context: {
+    'symbol': str,
+    'candles': deque[200],      # Recent OHLCV candles
+    'features': array,          # Extracted features
+    'last_update': datetime
+}
+
+# For prediction
+current_price: float
+now: datetime
+```
+
+### Outputs
+```python
+# Detected extrema
+extrema: {
+    'type': str,                # 'high' or 'low'
+    'price': float,
+    'timestamp': datetime,
+    'confidence': float,        # 0-1
+    'window_size': int
+}
+
+# Predicted pivot
+predicted_pivot: {
+    'type': str,                # 'high' or 'low'
+    'price': float,             # Predicted price level
+    'timestamp': datetime,      # Predicted time
+    'confidence': float,        # 0-1
+    'horizon_seconds': int      # Time until pivot (30-300s)
+}
+```
+
+### Training Data
+```python
+# Historical extrema for validation
+historical_extrema: List[{
+    'price': float,
+    'timestamp': datetime,
+    'type': str,
+    'detected': bool
+}]
+```
+
+---
+
+## Common Patterns
+
+### Action Encoding (All Models)
+```python
+0 = HOLD      # No action / maintain position
+1 = BUY       # Enter long / close short
+2 = SELL      # Enter short / close long
+```
+
+### Confidence Scores
+- Range: `0.0` to `1.0`
+- Typical threshold: `0.6` (60%)
+- High confidence: `> 0.8`
+- Low confidence: `< 0.4`
+
+### Batch Sizes
+- **Training**: Usually `1` (annotation-based) or `32-128` (batch training)
+- **Inference**: Usually `1` (real-time prediction)
+
+### Device Management
+All models support:
+- CPU: `torch.device('cpu')`
+- CUDA: `torch.device('cuda')`
+- Automatic device selection based on availability
+
+---
+
+## Model Selection Guide
+
+| Use Case | Recommended Model | Why |
+|----------|------------------|-----|
+| Multi-timeframe analysis | **Transformer** | Handles 150-candle sequences across timeframes |
+| Pattern recognition | **CNN** | Excellent at visual pattern detection |
+| Sequential decisions | **DQN** | Learns optimal action sequences via RL |
+| Order book dynamics | **COB RL** | Specialized for bid/ask imbalance |
+| Pivot detection | **Extrema** | Lightweight, fast pivot predictions |
+
+---
+
+## Integration Example
+
+```python
+# Get base data input
+base_input = data_provider.get_base_data_input(symbol, timestamp)
+
+# CNN prediction
+cnn_features = base_input.get_feature_vector()
+cnn_output = cnn_model(cnn_features)
+cnn_action = torch.argmax(cnn_output['action_probs'])
+
+# Transformer prediction
+transformer_batch = prepare_transformer_batch(base_input)
+transformer_output = transformer_model(**transformer_batch)
+transformer_action = torch.argmax(transformer_output['action_probs'])
+
+# DQN prediction
+dqn_state = base_input.get_feature_vector()
+dqn_output = dqn_agent.select_action(dqn_state)
+dqn_action = dqn_output['action']
+
+# Ensemble decision
+final_action = majority_vote([cnn_action, transformer_action, dqn_action])
+```
+
+---
+
+## Notes
+
+1. **Shape Conventions**: `[batch, ...]` indicates batch dimension first
+2. **Dtype**: All tensors use `torch.float32` unless specified
+3. **Gradients**: Only training targets require gradients
+4. **Normalization**: Features are typically normalized to `[-1, 1]` or `[0, 1]`
+5. **Missing Data**: Padded with zeros or last known values