# ADR-001: Data Processing and Aggregation Refactor

## Status
**Accepted**

## Context
The initial data collection and processing system was tightly coupled with the OKX exchange implementation. This made it difficult to add new exchanges, maintain the code, and ensure consistent data aggregation across different sources. Key issues included:
- Business logic mixed with data fetching.
- Inconsistent timestamp handling.
- No clear strategy for handling sparse data, leading to potential future data leakage.

A refactor was necessary to create a modular, extensible, and robust data processing pipeline that aligns with industry standards.

## Decision
We will refactor the data processing system to adhere to the following principles:

1.  **Modular & Extensible Design**: Separate exchange-specific logic from the core aggregation and storage logic using a factory pattern and base classes.
2.  **Right-Aligned Timestamps**: Adopt the industry standard for OHLCV candles where the timestamp represents the closing time of the interval. This ensures compatibility with major exchanges and historical data providers.
3.  **Sparse Candle Aggregation**: Emit candles only when trading activity occurs within a time bucket. This accurately reflects market activity and reduces storage.
4.  **No Future Leakage**: Implement a robust aggregation mechanism that only finalizes candles when their time period has definitively passed, preventing lookahead bias.
5.  **Centralized Repository for Database Operations**: Abstract all database interactions into a `Repository` pattern to decouple business logic from data persistence.

## Consequences

### Positive
- **Improved Maintainability**: Code is cleaner, more organized, and easier to understand.
- **Enhanced Extensibility**: Adding new exchanges is significantly easier.
- **Data Integrity**: Standardized timestamping and aggregation prevent data inconsistencies and lookahead bias.
- **Efficiency**: The sparse candle approach reduces storage and processing overhead.
- **Testability**: Decoupled components are easier to unit test.

### Negative
- **Initial Development Overhead**: The refactor required an initial time investment to design and implement the new architecture.
- **Increased Complexity**: The new system has more moving parts (factories, repositories), which may have a slightly steeper learning curve for new developers.

## Alternatives Considered

1.  **Keep the Monolithic Design**: Continue with the tightly coupled approach.
    - **Reason for Rejection**: This was not scalable and would have led to significant technical debt as new exchanges were added.
2.  **Use a Third-Party Data Library**: Integrate a library like `ccxt` for data collection.
    - **Reason for Rejection**: While powerful, these libraries did not offer the fine-grained control over the real-time aggregation and WebSocket handling that was required. Building a custom solution provides more flexibility.

## Related Documentation
- **Aggregation Strategy**: [docs/reference/aggregation-strategy.md](../reference/aggregation-strategy.md)
- **Data Collectors**: [docs/modules/data_collectors.md](../modules/data_collectors.md)
- **Database Operations**: [docs/modules/database_operations.md](../modules/database_operations.md)

---
*Back to [All Decisions (`./`)]*

# Refactored Data Processing Architecture

## Overview

The data processing system has been significantly refactored to improve reusability, maintainability, and scalability across different exchanges. The key improvement is the extraction of common utilities into a shared framework while keeping exchange-specific components focused and minimal.

## Architecture Changes

### Before (Monolithic)
```
data/exchanges/okx/
├── data_processor.py  # 1343 lines - everything in one file
├── collector.py
└── websocket.py
```

### After (Modular)
```
data/
├── common/                     # Shared utilities for all exchanges
│   ├── __init__.py
│   ├── data_types.py          # StandardizedTrade, OHLCVCandle, etc.
│   ├── aggregation.py         # TimeframeBucket, RealTimeCandleProcessor
│   ├── transformation.py      # BaseDataTransformer, UnifiedDataTransformer
│   └── validation.py          # BaseDataValidator, common validation
└── exchanges/
    └── okx/
        ├── data_processor.py  # ~600 lines - OKX-specific only
        ├── collector.py       # Updated to use common utilities
        └── websocket.py
```

## Key Benefits

### 1. **Reusability Across Exchanges**
- Candle aggregation logic works for any exchange
- Standardized data formats enable uniform processing
- Base classes provide common patterns for new exchanges

### 2. **Maintainability**
- Smaller, focused files are easier to understand and modify
- Common utilities are tested once and reused everywhere
- Clear separation of concerns

### 3. **Extensibility**
- Adding new exchanges requires minimal code
- New data types and timeframes are automatically supported
- Validation and transformation patterns are consistent

### 4. **Performance**
- Optimized aggregation algorithms and memory usage
- Efficient candle bucketing algorithms
- Lazy evaluation where possible

### 5. **Testing**
- Modular components are easier to test independently

## Time Aggregation Strategy

### Right-Aligned Timestamps (Industry Standard)

The system uses **RIGHT-ALIGNED timestamps** following industry standards from major exchanges (Binance, OKX, Coinbase):

- **Candle timestamp = end time of the interval (close time)**
- 5-minute candle with timestamp `09:05:00` represents data from `09:00:01` to `09:05:00`
- 1-minute candle with timestamp `14:32:00` represents data from `14:31:01` to `14:32:00`
- This aligns with how exchanges report historical data

### Aggregation Process (No Future Leakage)

```python
def process_trade_realtime(trade: StandardizedTrade, timeframe: str):
    """
    Real-time aggregation with strict future leakage prevention
    
    CRITICAL: Only emit completed candles, never incomplete ones
    """
    
    # 1. Calculate which time bucket this trade belongs to
    trade_bucket_start = get_bucket_start_time(trade.timestamp, timeframe)
    
    # 2. Check if current bucket exists and matches
    current_bucket = current_buckets.get(timeframe)
    
    # 3. Handle time boundary crossing
    if current_bucket is None:
        # First bucket for this timeframe
        current_bucket = create_bucket(trade_bucket_start, timeframe)
    elif current_bucket.start_time != trade_bucket_start:
        # Time boundary crossed - complete previous bucket FIRST
        if current_bucket.has_trades():
            completed_candle = current_bucket.to_candle(is_complete=True)
            emit_candle(completed_candle)  # Store in market_data table
        
        # Create new bucket for current time period
        current_bucket = create_bucket(trade_bucket_start, timeframe)
    
    # 4. Add trade to current bucket
    current_bucket.add_trade(trade)
    
    # 5. Return only completed candles (never incomplete/future data)
    return completed_candles  # Empty list unless boundary crossed
```

### Time Bucket Calculation Examples

```python
# 5-minute timeframes (00:00, 00:05, 00:10, 00:15, etc.)
trade_time = "09:03:45"  -> bucket_start = "09:00:00", bucket_end = "09:05:00"
trade_time = "09:07:23"  -> bucket_start = "09:05:00", bucket_end = "09:10:00"
trade_time = "09:05:00"  -> bucket_start = "09:05:00", bucket_end = "09:10:00"

# 1-hour timeframes (align to hour boundaries)
trade_time = "14:35:22"  -> bucket_start = "14:00:00", bucket_end = "15:00:00"
trade_time = "15:00:00"  -> bucket_start = "15:00:00", bucket_end = "16:00:00"

# 4-hour timeframes (00:00, 04:00, 08:00, 12:00, 16:00, 20:00)
trade_time = "13:45:12"  -> bucket_start = "12:00:00", bucket_end = "16:00:00"
trade_time = "16:00:01"  -> bucket_start = "16:00:00", bucket_end = "20:00:00"
```

### Future Leakage Prevention

**CRITICAL SAFEGUARDS:**

1. **Boundary Crossing Detection**: Only complete candles when trade timestamp definitively crosses time boundary
2. **No Premature Completion**: Never emit incomplete candles during real-time processing
3. **Strict Time Validation**: Trades only added to buckets if `start_time <= trade.timestamp < end_time`
4. **Historical Consistency**: Same logic for real-time and historical processing

```python
# CORRECT: Only complete candle when boundary is crossed
if current_bucket.start_time != trade_bucket_start:
    # Time boundary definitely crossed - safe to complete
    completed_candle = current_bucket.to_candle(is_complete=True)
    emit_to_storage(completed_candle)

# INCORRECT: Would cause future leakage
if some_timer_expires():
    # Never complete based on timers or external events
    completed_candle = current_bucket.to_candle(is_complete=True)  # WRONG!
```

### Data Storage Flow

```
WebSocket Trade Data → Validation → Transformation → Aggregation → Storage
     |                                                      |            |
     ↓                                                      ↓            ↓
Raw individual trades                              Completed OHLCV    Incomplete OHLCV
     |                                              candles (storage)  (monitoring only)
     ↓                                                      |
raw_trades table                                   market_data table
(debugging/compliance)                             (trading decisions)
```

**Storage Rules:**
- **Raw trades** → `raw_trades` table (every individual trade/orderbook/ticker)
- **Completed candles** → `market_data` table (only when timeframe boundary crossed)
- **Incomplete candles** → Memory only (never stored, used for monitoring)

### Aggregation Logic Implementation

```python
def aggregate_to_timeframe(trades: List[StandardizedTrade], timeframe: str) -> List[OHLCVCandle]:
    """
    Aggregate trades to specified timeframe with right-aligned timestamps
    """
    # Group trades by time intervals
    buckets = {}
    completed_candles = []
    
    for trade in sorted(trades, key=lambda t: t.timestamp):
        # Calculate bucket start time (left boundary)
        bucket_start = get_bucket_start_time(trade.timestamp, timeframe)
        
        # Get or create bucket
        if bucket_start not in buckets:
            buckets[bucket_start] = TimeframeBucket(timeframe, bucket_start)
        
        # Add trade to bucket
        buckets[bucket_start].add_trade(trade)
    
    # Convert all buckets to candles with right-aligned timestamps
    for bucket in buckets.values():
        candle = bucket.to_candle(is_complete=True)
        # candle.timestamp = bucket.end_time (right-aligned)
        completed_candles.append(candle)
    
    return completed_candles
```

## Common Components

### Data Types (`data/common/data_types.py`)

**StandardizedTrade**: Universal trade format
```python
@dataclass
class StandardizedTrade:
    symbol: str
    trade_id: str
    price: Decimal
    size: Decimal
    side: str  # 'buy' or 'sell'
    timestamp: datetime
    exchange: str = "okx"
    raw_data: Optional[Dict[str, Any]] = None
```

**OHLCVCandle**: Universal candle format
```python
@dataclass
class OHLCVCandle:
    symbol: str
    timeframe: str
    start_time: datetime
    end_time: datetime
    open: Decimal
    high: Decimal
    low: Decimal
    close: Decimal
    volume: Decimal
    trade_count: int
    is_complete: bool = False
```

### Aggregation (`data/common/aggregation.py`)

**RealTimeCandleProcessor**: Handles real-time candle building for any exchange
- Processes trades immediately as they arrive
- Supports multiple timeframes simultaneously
- Emits completed candles when time boundaries cross
- Thread-safe and memory efficient

**BatchCandleProcessor**: Handles historical data processing
- Processes large batches of trades efficiently
- Memory-optimized for backfill scenarios
- Same candle output format as real-time processor

### Transformation (`data/common/transformation.py`)

**BaseDataTransformer**: Abstract base class for exchange transformers
- Common transformation utilities (timestamp conversion, decimal handling)
- Abstract methods for exchange-specific transformations
- Consistent error handling patterns

**UnifiedDataTransformer**: Unified interface for all transformation scenarios
- Works with real-time, historical, and backfill data
- Handles batch processing efficiently
- Integrates with aggregation components

### Validation (`data/common/validation.py`)

**BaseDataValidator**: Common validation patterns
- Price, size, volume validation
- Timestamp validation
- Orderbook validation
- Generic symbol validation

## Exchange-Specific Components

### OKX Data Processor (`data/exchanges/okx/data_processor.py`)

Now focused only on OKX-specific functionality:

**OKXDataValidator**: Extends BaseDataValidator
- OKX-specific symbol patterns (BTC-USDT format)
- OKX message structure validation
- OKX field mappings and requirements

**OKXDataTransformer**: Extends BaseDataTransformer
- OKX WebSocket format transformation
- OKX-specific field extraction
- Integration with common utilities

**OKXDataProcessor**: Main processor using common framework
- Uses common validation and transformation utilities
- Significantly simplified (~600 lines vs 1343 lines)
- Better separation of concerns

### Updated OKX Collector (`data/exchanges/okx/collector.py`)

**Key improvements:**
- Uses OKXDataProcessor with common utilities
- Automatic candle generation for trades
- Simplified message processing
- Better error handling and statistics
- Callback system for real-time data

## Usage Examples

### Creating a New Exchange

To add support for a new exchange (e.g., Binance):

1. **Create exchange-specific validator:**
```python
class BinanceDataValidator(BaseDataValidator):
    def __init__(self, component_name="binance_validator"):
        super().__init__("binance", component_name)
        self._symbol_pattern = re.compile(r'^[A-Z]+[A-Z]+$')  # BTCUSDT format
    
    def validate_symbol_format(self, symbol: str) -> ValidationResult:
        # Binance-specific symbol validation
        pass
```

2. **Create exchange-specific transformer:**
```python
class BinanceDataTransformer(BaseDataTransformer):
    def transform_trade_data(self, raw_data: Dict[str, Any], symbol: str) -> Optional[StandardizedTrade]:
        return create_standardized_trade(
            symbol=raw_data['s'],          # Binance field mapping
            trade_id=raw_data['t'],
            price=raw_data['p'],
            size=raw_data['q'],
            side='buy' if raw_data['m'] else 'sell',
            timestamp=raw_data['T'],
            exchange="binance",
            raw_data=raw_data
        )
```

3. **Automatic candle support:**
```python
# Real-time candles work automatically
processor = RealTimeCandleProcessor(symbol, "binance", config)
for trade in trades:
    completed_candles = processor.process_trade(trade)
```

### Using Common Utilities

**Data transformation:**
```python
# Works with any exchange
transformer = UnifiedDataTransformer(exchange_transformer)
standardized_trade = transformer.transform_trade_data(raw_trade, symbol)

# Batch processing
candles = transformer.process_trades_to_candles(
    trades_iterator, 
    ['1m', '5m', '1h'], 
    symbol
)
```

**Real-time candle processing:**
```python
# Example usage
from data.common.aggregation.realtime import RealTimeCandleProcessor

processor = RealTimeCandleProcessor(symbol, "binance", config)
processor.add_candle_callback(on_candle_completed)
processor.process_trade(trade)
```

```python
# Example usage
from data.common.aggregation.realtime import RealTimeCandleProcessor

candle_processor = RealTimeCandleProcessor(symbol, exchange, config)
candle_processor.add_candle_callback(on_candle_completed)
candle_processor.process_trade(trade)
```

## Testing

The refactored architecture includes comprehensive testing:

**Test script:** `scripts/test_refactored_okx.py`
- Tests common utilities
- Tests OKX-specific components
- Tests integration between components
- Performance and memory testing

**Run tests:**
```bash
python scripts/test_refactored_okx.py
```

## Migration Guide

### For Existing OKX Code

1. **Update imports:**
```python
# Old
from data.exchanges.okx.data_processor import StandardizedTrade, OHLCVCandle

# New
from data.common import StandardizedTrade, OHLCVCandle
```

2. **Use new processor:**
```python
# Old
from data.exchanges.okx.data_processor import OKXDataProcessor, UnifiedDataTransformer

# New
from data.exchanges.okx.data_processor import OKXDataProcessor  # Uses common utilities internally
```

3. **Existing functionality preserved:**
- All existing APIs remain the same
- Performance improved due to optimizations
- More features available (better candle processing, validation)

### For New Exchange Development

1. **Start with common base classes**
2. **Implement only exchange-specific validation and transformation**
3. **Get candle processing, batch processing, and validation for free**
4. **Focus on exchange API integration rather than data processing logic**

## Performance Improvements

**Memory Usage:**
- Streaming processing reduces memory footprint
- Efficient candle bucketing algorithms
- Lazy evaluation where possible

**Processing Speed:**
- Optimized validation with early returns
- Batch processing capabilities
- Parallel processing support

**Maintainability:**
- Smaller, focused components
- Better test coverage
- Clear error handling and logging

## Future Enhancements

**Planned Features:**
1. **Exchange Factory Pattern** - Automatically create collectors for any exchange
2. **Plugin System** - Load exchange implementations dynamically
3. **Configuration-Driven Development** - Define new exchanges via config files
4. **Enhanced Analytics** - Built-in technical indicators and statistics
5. **Multi-Exchange Arbitrage** - Cross-exchange data synchronization

This refactored architecture provides a solid foundation for scalable, maintainable cryptocurrency data processing across any number of exchanges while keeping exchange-specific code minimal and focused.