feat: Phase 1 - Synthetic Data Training Integration

[hudsonaikins]

Summary

This PR implements Phase 1 of the synthetic data training integration, establishing the foundation for training Kalshi agents using synthetic data while maintaining full compatibility with the production Redis infrastructure.

Key Components Implemented

1. Synthetic Data Injector (src/integration/synthetic_injector.py)

• Injects synthetic game and trading data into Redis channels
• Multiple timing strategies: realtime, accelerated, burst, adaptive
• Market reaction generation with price impact modeling
• Performance metrics tracking

2. Training Bridge (src/integration/training_bridge.py)

• Central coordinator between synthetic data and agent training
• Configurable training modes: exploration, exploitation, validation, production_prep
• Integration with AgentAnalytics, MemorySystem, and ConfidenceCalibrator
• Kelly Criterion calculations and decision tracking

3. Agent Training Harness (src/integration/training_harness.py)

• Orchestrates complete end-to-end training sessions
• 10 pre-defined training scenarios
• Adaptive difficulty adjustment
• Real-time monitoring with checkpointing
• Comprehensive metrics: Sharpe ratio, win rate, P&L

4. Training Consumer (src/agents/training_consumer.py)

• Extended BaseAgentRedisConsumer with training capabilities
• Decision tracking with context preservation
• Confidence calibration integration
• Memory system hooks

Testing

• ✅ All modules import successfully
• ✅ Syntax validation passed
• ✅ Import path corrections applied

Next Steps (Future PRs)

• Phase 2: Decision Tracking Middleware
• Phase 3: Training Mode Controller
• Phase 4: Performance optimization and benchmarking

Notes

This establishes the core infrastructure needed for synthetic data training. The system can now train agents offline using generated data, reducing API costs while maintaining production compatibility.