CloneTrader

Mission

CTO.new Project Brief: The Systematic Trader Cloning Engine (STC-E) 1. Project Objective Build an agentic AI system that observes a professional manual trader, deconstructs their intuition into a probabilistic risk model, and deploys an autonomous replica onto MT5 or cTrader. The goal is to solve the "black box" problem of standard EAs by creating adaptive systems that learn how a trader thinks, not just what they buy. 2. Technical Architecture (The 3-Layer Clone) Layer 1: Observation & Deconstruction (The "Ghosting" Phase) · Input: Trade logs, screenshots, news reactions, and journal entries. · Mechanism: Use alpha-skills suite modules like trade-hypothesis-ideator and signal-postmortem to reverse-engineer entries . · Output: A "Trader Personality JSON"—defining variables like Risk Aversion Volatility, Confirmation Bias Lag, and Max Adverse Excursion (MAE) tolerance. Layer 2: Learning & Validation · Tech Stack: Python (Scikit-learn/TensorFlow) for classification. · Process: Train a classifier (XGBoost or Neural Network) to predict the trader's decision (Buy/Sell/Pass) based on market microstructure . · Validation: Walk-forward analysis using TimeSeriesSplit to ensure the AI isn't overfitting to noise . Layer 3: Autonomous Deployment (The "Clone") · Platforms: Export via ONNX runtime for MT5 (MQL5) or compile as a cBot for cTrader (C#). · Execution Logic: The EA does not use fixed indicators; it runs the exported neural net on every tick to replicate the probability distribution of the human's actions . 3. Core Features Feature Implementation Detail Multi-Platform Bridge Python connector pulls data from MT5/cTrader; trained model pushes decisions back via API or DLL . Regime Detection Before trading, the system checks macro-regime-detector to see if the current market matches the trader's "comfort zone" . Semantic Stop-Loss Uses thesis invalidation logic (e.g., "If VIX > Trader's Historical Max, halt clone") rather than fixed pips . Hallucination Guard AI code often has errors (e.g., bad lot calculations). The system includes a CI/CD for Trading—auto-backtesting and rejecting code that violates drawdown limits . 4. Why Traditional Copy Trading Fails (And This Wins) Standard copy trading (cTrader Copy, Mirror Trading) simply duplicates lots . If the professional takes a day off or faces a black swan, the follower bleeds. This project creates a "Digital Twin" : The AI learns to abort trades the human would abort and scale trades the human scales. It replicates the conditioning behind the trade . 5. Deliverables (CTO.new Output) 1. The Analyzer: A Python script that ingests a human's trade history and outputs a "Skill Fingerprint." 2. The Adapter: A script that converts that fingerprint into an ONNX neural net. 3. The Robot: Native source code (.mq5 for MT5 or .cs for cTrader) ready for deployment . 4. The Dashboard: Streamlit view showing "Clone Confidence %" (how closely the AI matches the human's historical decisions). 6. Success Metrics · R-Squared (Action): Statistical similarity between human and AI decision sequence on Out-of-Sample (OOS) data. · Sharpe Ratio Retention: The AI must retain at least 80% of the human's Sharpe ratio after transaction costs.