GNN-PPO Python AI Engine Connected to MT4 Non-Repaint Arrow Indicator

1. SYSTEM ARCHITECTURE:

- Python AI Backend: A Graph Neural Network (PyTorch Geometric) + PPO (Stable-Baselines3) model analyzing 6-8 correlated pairs simultaneously (EURUSD, GBPUSD, USDJPY, USDCHF, AUDUSD, USDCAD, XAUUSD, DXY) on the M1 timeframe.

- Action Space & Reward: Outputs CALL, PUT, or HOLD for a 5-minute expiration (5 M1 candles). Trained with payout-adjusted reward shaping (+0.80 win / -1.00 loss) using Walk-Forward Validation on exported MT4 M1 historical data.

- Communication Bridge: Local TCP Socket or ZeroMQ (<100ms total latency) connecting MT4 and Python.

2. MT4 CUSTOM INDICATOR (FRONTEND):

- Synchronizes M1 candle timestamps across all selected pairs and sends data to Python.

- Draws permanent, strictly NON-REPAINTING Green UP (CALL) and Red DOWN (PUT) arrows within 500ms of the M1 candle close.

- Includes audio/pop-up alerts, optional Telegram notification, confidence threshold input parameter, and an on-chart dashboard showing connection status and historical Win/Loss rate (evaluated at the 5th M1 candle close).

3. DELIVERABLES (MANDATORY):

- Full unlocked source codes (.py scripts and .mq4 file + compiled .ex4).

- Trained model weights (.pt or .onnx) + retraining script.

- Step-by-step installation guide (requirements.txt, MT4 DLL/WebRequest settings).

- Out-of-Sample backtest report (minimum 3 months of real market data).

SCREENING QUESTIONS (Must be answered in your application):

1. How will you synchronize M1 candle timestamps across 6+ pairs before feeding the graph to the GNN to prevent data leakage?

2. Have you previously connected Python ML models to MT4 via ZeroMQ or Sockets? Briefly explain your approach.