DareFX is an automated trading terminal and daily swing execution system for EUR/USD. It combines supervised machine learning models with volatility-based barrier exits and automated position management over MetaTrader 5.
The system runs daily after the New York close (17:00 EST / 22:00 UTC), generates directional predictions using walk-forward trained LightGBM models, calculates fixed-risk position sizes, and executes orders directly via MetaTrader 5 with active break-even trailing protection.
Standalone desktop binaries and installers are available directly from GitHub Releases:
| Platform | Format | Download Link | Size | Details |
|---|---|---|---|---|
| Windows (x86_64) | .exe |
DareFX-windows-x86_64.exe | 186 MB | Portable executable with native MetaTrader 5 API integration |
| Linux (Debian / Ubuntu / Mint) | .deb |
darefx_1.0.0_amd64.deb | 526 MB | Standard Debian package with desktop launcher and app menu entry |
| Linux (Generic x86_64) | .tar.gz |
DareFX-linux-x86_64.tar.gz | 534 MB | Self-contained binary archive |
# Windows
# Download DareFX-windows-x86_64.exe and double-click to launch
# Linux (Debian / Ubuntu / Linux Mint)
sudo dpkg -i darefx_1.0.0_amd64.deb
darefx-gui
# Linux (Portable Archive)
tar -xzf DareFX-linux-x86_64.tar.gz
./DareFXThe desktop client (app.py) is built with PySide6 / Qt and connects directly to the local or remote MT5 bridge:
- Balance, equity, margin, and floating PnL summary cards
- Active position table with live pip tracking, break-even status indicators, and individual close buttons
- Controls for auto-trading, panic close, manual order entry, and on-demand model inference
- System tray integration for background monitoring without interrupting open trades
# Launch GUI from source
python app.py
# Or on Windows using the batch launcher
run_gui.bat- Feature pipeline: Computes 35 stationary indicators (RSI-14, MACD, Bollinger Bands, ATR-14, NATR, ADX-14, EMA ratios) normalized with rolling backward-looking z-scores to prevent data leakage.
- Labeling and exits: Uses triple-barrier labeling where profit targets and stop losses scale with rolling ATR, alongside a 5-day maximum holding period.
- Validation: Walk-forward cross-validation across 79 folds (36 months train, 6 months test, 3 months step) with a 10-day purge gap and 5-day embargo to eliminate overlap bias.
- Risk controls: Fixed fractional position sizing (1.0% equity risk per trade) capped by half-Kelly constraints, automatic break-even adjustment (+1.5 pips profit lock once price advances +15 pips), and circuit breakers for daily drawdown limits.
- MetaTrader 5 bridge: Connects directly on Windows to the MT5 terminal API, or remotely from Linux through a secure local bridge tunnel.
Market Data (Daily OHLCV)
│
▼
Feature Pipeline (35 Indicators, Rolling Z-Score)
│
▼
LightGBM Walk-Forward Inference ──► ADX Trend Filter (ADX >= 32.0)
│
▼
Position Sizer (1% Risk / Half-Kelly)
│
▼
MT5 Execution Bridge ──► Break-Even Trailing Guard (+1.5 pips lock)
Evaluated across 4,519 daily bars under realistic transaction friction (1.5 pips spread, 1.0 pip slippage, and $7.00/lot commission):
| Metric | Result | Benchmark | Status |
|---|---|---|---|
| Total Net Return | +37.58% ($10k → $13,757) | > 0.00% | Pass |
| Win Rate | 59.02% (72 wins / 50 losses) | > 50.00% | Pass |
| Profit Factor | 1.72 | > 1.30 | Pass |
| Annualized Sharpe Ratio | 4.24 | > 0.50 | Pass |
| Probabilistic Sharpe Ratio (PSR) | 99.7% | > 95.0% | Pass |
| Deflated Sharpe Ratio (DSR) | 98.2% (35 trials) | > 50.0% | Pass |
| Maximum Drawdown | 4.62% | < 15.00% | Pass |
| Trade Expectancy | +$30.80 / trade | > $0.00 | Pass |
Full evaluation details are available in docs/EVALUATION-REPORT.md.
- Python 3.12
- uv (recommended) or pip
- MetaTrader 5 terminal (for live execution)
# Clone the repository
git clone https://github.com/davinmaritza/darefx.git
cd darefx
# Set up virtual environment and install dependencies
uv venv --python 3.12 .venv
source .venv/bin/activate
uv pip install -e ".[dev]"
# Run test suite
pytest -vDareFX provides a command-line interface for the data pipeline, training, and execution:
# Fetch daily EUR/USD bars into annual Parquet partitions
darefx fetch --source yahoo --symbol EURUSD --range-str 25ydarefx features --symbol EURUSD --pt 1.5 --sl 1.5 --holding 5darefx train --symbol EURUSD --train-months 36 --test-months 6 --step-months 3darefx backtest --symbol EURUSD --threshold 0.46 --adx-min 32.0 --sl 1.0 --pt 1.2darefx evaluate --symbol EURUSD --trials 35# Simulation / Dry-run
darefx run-cycle --symbol EURUSD --equity 10000
# Live execution via MT5 Bridge
darefx run-cycle --symbol EURUSD --equity 10000 --executedarefx risk-statusARCHITECTURE.md— Component architecture and data flow diagramsRISK-MANAGEMENT.md— Sizing models, break-even mechanics, and drawdown limitsDATA-PIPELINE.md— Ingestion, tick cleaning, resampling, and Parquet storageDEPLOYMENT-RUNBOOK.md— Systemd service and production deployment guideEVALUATION-REPORT.md— Out-of-sample backtest report and statistical audit
This project is licensed under the MIT License. See LICENSE for details.