From trend-following to AI-driven quantitative models — choose your systematic edge
The aggressor. While others take one entry and pray, v9 scales into winners with Fibonacci pyramid entries at 38.2%, 50%, and 61.8% retracement levels. When the market gives you a pullback, v9 doesn't hesitate — it adds. AutoTune engine continuously recalibrates parameters based on live performance.
The Fibonacci Pyramid uses 3-level scaling at golden ratio levels (38.2%, 50%, 61.8%). Entry conditions require a minimum of 7/9 confluence factors including Smart Money Concepts (BOS, CHoCH, Order Blocks, FVG), Nadaraya-Watson kernel regression, and ELM neural filtering.
AutoTune recalibrates EMA periods, ATR multipliers, RSI thresholds, and trailing stop distances every N bars. Walk-forward validation prevents overfitting. Compatible with TradingView and Pine Script v5/v6.
The optimizer. Each of the 1,845 instruments gets its own mathematically tuned parameter set through Bayesian optimization (100 Optuna trials per ticker). Then the 4-layer AI scoring engine filters: FinRL PPO (35%), FinBERT sentiment (20%), technical indicators (25%), and Claude AI reasoning (20%). Only signals scoring 65+ pass through.
PerSymbol AutoTune runs Bayesian optimization via Optuna's TPE sampler across 31+ parameters: EMA periods (fast/slow/base), ATR multipliers, SL/TP ratios, RSI overbought/oversold thresholds, Bollinger Band periods and deviations, trailing stop activation, regime detection windows, and position sizing. Each ticker receives 100 optimization trials with walk-forward validation (70% train / 30% test) to prevent curve fitting.
The brain. FinRL's Proximal Policy Optimization (PPO) agent trained on 847,000 episodes across 1,796 tickers. The agent observes 14 technical and fundamental factors simultaneously, learning optimal entry/exit timing and position sizing. This is the same model powering our live $100K+ portfolio.
The FinRL PPO agent uses a 3-layer neural network (256-128-64) with dropout regularization. Training environment simulates realistic market conditions with transaction costs, slippage, and partial fills. The 14 confluence factors include: RSI, MACD histogram, Bollinger Band position, ATR-normalized volatility, OBV trend, ADX strength, Stochastic oscillator, VWAP deviation, EMA crossover state, volume profile, FinBERT sentiment score, Claude reasoning output, market regime classification, and momentum score. The agent was trained on 5 years of historical data (2019-2024) with out-of-sample validation on 2025 data achieving a 61.2% win rate.
| Feature | v9 Fibonacci | v10 AutoTune | v11 FinRL |
|---|---|---|---|
| Approach | Fibonacci scaling | Bayesian optimization | Reinforcement learning |
| AI Integration | — | 4-layer scoring | Full PPO agent |
| FinBERT Sentiment | — | ✓ | ✓ |
| Per-Symbol Optimization | 200 presets | 1,845 presets | 1,796 trained |
| Source Format | Pine Script | Pine Script + JSON | Python + Model |
| Best For | TradingView users | Multi-asset traders | Quant developers |
| Price | $199 | $299 | $399 |