Lorentzian-RLS Adaptive Liquidity Engine

Family: hybrid · Regime: trending · Complexity: high · Asset classes: Equities, Forex, Crypto · Timeframes: 1H, 4H, Daily

Thesis

Market edges exist where institutional liquidity zones (Order Blocks) align with non-linear trend clusters (Lorentzian Classification). By filtering these signals through an adaptive RLS FIR filter, we capture moves where momentum and structure converge, while using MAPE to scale risk according to the model's current predictive failure rate.

Components

• UDT Array Push Identity Test (regime) — Serves as a data integrity filter; ensures that multi-type data structures (Close/High arrays) are synchronized before allowing execution.
• PineForge demo (direction) — Provides the raw price overlay (y=close) used as the baseline for all distance-to-level calculations.
• Lorentzian Classification (entry) — Uses non-Euclidean distance to identify historical price clusters that preceded profitable moves, providing the core predictive signal.
• Q-Levels V2.2 (exit) — Provides structural exits based on external gamma walls or supply/demand levels parsed from custom text strings.
• Mean Absolute Percentage Error (MAPE) (risk) — Quantifies the model's prediction error volatility to dynamically scale stop-loss distances and position size.
• Order Block & FVG Detector (confirmation) — Validates ML entries by requiring institutional liquidity presence (Order Blocks) to confirm a change in market structure.
• Recursive Least Squares Adaptive Filter (RLS) (volatility_filter) — Acts as a denoising filter; ensures that entries only occur when the adaptive trend is aligned with the ML prediction.

Known failure conditions

• Q-Levels string data becomes stale or incorrectly formatted, leading to missing exit targets.
• Lorentzian Classification encounters 'curse of dimensionality' where feature variance renders KNN voting random.
• MAPE produces division-by-zero errors in assets approaching zero value (e.g., distressed equities/shitcoins).

Explore the full interactive blueprint with parameter ranges and evidence on WOBR StrategyVerse, or generate this strategy as an MT4/MT5 Expert Advisor with QuantMogul AI Engine (free download).