All notable changes to this project will be documented in this file.

• Sanitized public release documentation without private operational reports or operator-specific model assignments.
• v2.5 release checklist in RELEASE_CHECKLIST.md.
• Script release-surface index in scripts/README.md.
• English NanoFenix HTF v2.5 documentation in docs/NANOFENIX_HTF_V2_5_CHANGES.md.
• Hybrid runner support for staged TP/SL tracking and partial closes.
• Binance Futures algorithmic-order checks when verifying live protective orders.

• README and quick-start documentation updated for the official v2.5.0 release.
• NanoFenix changelog has been translated to English.
• MiniFenix README has been translated to English and reframed as a two-speed research prototype.
• NanoFenix adaptive-fusion comments and docstrings have been translated to English.
• Hybrid runner comments and docstrings around staged exits have been translated to English.
• Live engine restart behavior hydrates an existing exchange position before assuming the bot is flat.
• Live slot summaries now account for hydrated exchange positions when validating close events.
• Paper/hybrid execution guards reject invalid prices before recording position transitions.
• Live suite plans can now carry explicit lite/MTF guard settings into run_fenix_live_slot.py, avoiding hidden dependence on inherited shell variables.
• NanoFenix trailing stops account for estimated fees before treating an exit as acceptable.

• Failed execution attempts are no longer recorded as realized trading losses by runtime risk accounting.
• Protective-order verification no longer misses open Binance Futures algorithmic orders.
• A safe reduce-only close of a hydrated live position no longer causes a false completed_with_accounting_gap summary.
• Paper/hybrid logs are less likely to contain misleading zero or missing entry prices.
• technical_mtf_qabba_guard now reports missing strict-MTF configuration instead of silently recording _mtf_bias: {}.
• A Technical timeout/error fallback can no longer count as a clean Technical HOLD for the optional MTF+QABBA entry path.

This release represents a complete rewrite of FenixAI, transitioning from CrewAI to LangGraph for a more robust and flexible agent orchestration system.

• LangGraph Orchestrator: Replaced CrewAI with LangGraph state machine for agent coordination
• Multi-Provider LLM System: Support for Ollama, MLX (Apple Silicon), Groq, and HuggingFace
• ReasoningBank: New memory system inspired by academic research (arXiv:2509.25140)
  • Semantic search for similar past decisions
  • Self-evaluation with LLM-as-Judge
  • Reward shaping based on trade outcomes
• Dynamic Weighting: Agents weights adapt based on historical performance and market conditions

• Enhanced Technical Analyst: Improved indicator validation and confluence scoring
• Enhanced Visual Analyst: Better chart pattern recognition with security validation
• Enhanced Sentiment Analyst: Multi-source sentiment aggregation
• Enhanced QABBA Agent: Bollinger Bands and volatility analysis
• Enhanced Decision Agent: Weighted consensus with LLM-as-Judge integration
• Risk Manager: Circuit breakers and dynamic position sizing

• New React Dashboard: Built with Vite, TypeScript, and TailwindCSS
• Real-time Updates: Socket.IO for live data streaming
• Agent Performance Charts: Visualize agent accuracy and contributions
• System Status Monitoring: CPU, memory, and connection health

• Async Binance Client: Improved WebSocket handling
• Paper Trading Mode: Realistic simulation with slippage and fees
• Multi-timeframe Analysis: Context from 15m, 1h, and 4h timeframes

• SecureSecretsManager: Encrypted storage for API keys
• Chart Path Validation: Prevents path traversal and stale data attacks
• Live Trading Safeguard: Requires explicit --allow-live flag

• Project Structure: Reorganized into src/ with clear module separation
• Configuration: YAML-based configuration with multiple LLM provider profiles
• Logging: Structured logging with JSON trace files for debugging
• Dependencies: Updated to Python 3.10+, modern async patterns

• CrewAI dependency (replaced by LangGraph)
• Old monolithic live_trading.py (split into modules)
• Deprecated agent implementations (moved to legacy/)

• New comprehensive documentation in /docs
• Legacy development notes preserved in /legacy/docs/v2_development_notes

• CrewAI-based Architecture: Multi-agent system using CrewAI
• Basic Agents: Technical, Visual, Sentiment, Consensus
• Binance Integration: Live and paper trading
• Simple Dashboard: Flask-based monitoring
• Paper Trading System: Basic simulation

1. Configuration Format

   • Old: config/config.yaml
   • New: config/fenix.yaml + config/llm_providers.yaml

2. Agent Imports

   • Old: from agents.sentiment import SentimentAgent
   • New: from src.agents.sentiment_enhanced import EnhancedSentimentAnalyst

3. Running the Bot

   • Old: python live_trading.py
   • New: python run_fenix.py --api

4. Environment Variables

   • New required variables for LLM providers
   • See .env.example for complete list

1. Backup your configuration and trades
2. Install new dependencies: pip install -e ".[dev]"
3. Update configuration files
4. Pull required Ollama models
5. Test with paper trading before going live

We use Semantic Versioning:

• MAJOR: Incompatible API changes
• MINOR: New features, backwards compatible
• PATCH: Bug fixes, backwards compatible

See CONTRIBUTING.md for guidelines on:

• Bug reports
• Feature requests
• Pull requests
• Code style