OmniScreener
Systematic cryptocurrency market analysis, execution simulation, and quantitative research engine.
01. Problem & Context
Quantitative traders and market researchers face high barriers when evaluating breakout signals: retail screener tools suffer from arbitrary rate limits, lack point-in-time historical data accuracy, and encourage data snooping through uncorrected multiple hypothesis testing. OmniScreener was engineered strictly as an empirical research and paper-trading simulation environment to test market hypothesis models without capital risk or live execution.
02. Engineering Solution
Constructed a high-performance desktop analytical workstation in Flutter/Dart that ingests Coinbase-tradable USD pairs, calculates real-time volatility contraction and relative volume (RVOL) ignition metrics, and subjects signals to an empirical statistical validation engine adhering to strict holdout isolation and family-wise error rate corrections. The system is strictly non-execution: live order placement is explicitly guarded, prohibited, and omitted.
03. System Architecture
Built around a deterministic, unidirectional data pipeline. Market candles are pulled via rate-throttled REST clients, normalized into chronological time-series buffers, and evaluated across eight orthogonal evidence dimensions. All statistical evaluations enforce strict separation between Discovery (50%), Validation (25%), and Holdout (25%) partitions.
Core Architectural Components
Market Ingestion & Throttling Buffer
Asynchronous ingestion engine managing Coinbase Exchange public endpoints with request batching, token-bucket throttling, and exponential backoff to ensure zero dropped candles.
Ignition & Volatility Scoring Pipeline
Mathematical engine computing price range expansion, close quartile position, rolling RVOL, and 1-hour range breakouts across 15m and 1h intervals.
Statistical Discovery & Holdout Engine
Validation layer isolating 25% chronological out-of-sample holdout partitions, completely frozen from parameter tuning to prevent data snooping and lookahead leakage.
Multi-Testing Correction Suite
Statistical auditors implementing Bonferroni FWER, Holm-Bonferroni step-down, Benjamini-Hochberg FDR (q=0.10), Sidak multi-trial deflation, and Bailey/López de Prado Deflated Sharpe Ratio (DSR).
Execution & Slippage Simulator
Paper-trading simulation engine modeling order-book maker execution drag, bid-ask spread costs, and latency decay curves across Bar 1 (T0) through Bar 4+ intervals.
04. Notable Engineering Decisions
Zero Live-Execution Guardrails
The system is strictly designed for quantitative research, backtesting, and paper execution simulation. To maintain rigorous safety standards, live order placement and cancellation pathways are intentionally guarded, prohibited, and omitted from the codebase.
Non-Parametric Bootstrap over Gaussian Assumptions
Financial return distributions exhibit fat tails, skewness, and non-normality. Standard Gaussian confidence intervals underestimate downside variance, so the system utilizes 1,000-iteration deterministic bootstrap resampling and two-tailed Wilson score intervals.
Early Detection Latency & Decay Tracking
Empirical testing proved that waiting for breakout confirmation beyond Bar 1 (T0) and Bar 2 (T1) sacrifices 28.1% of maximum favorable excursion (MFE) with 38.9% information decay. Tracking decay explicitly provides realistic expectancy estimates.
05. Engineering Challenges & Solutions
Challenge: Preventing lookahead bias in historical candle evaluation.
Engineering Solution: Engineered an adversarial outcome mutation test suite that systematically mutates future bar timestamps to verify that point-in-time feature extraction remains 100% invariant to subsequent price action.
Challenge: High-density UI rendering of multi-column statistical tables.
Engineering Solution: Leveraged Flutter desktop canvas rendering with virtualized list viewports and memoized Riverpod selectors, achieving smooth 60fps sorting across hundreds of instrument rows.
06. Implementation Highlights
- Rigorous test suite with 370+ automated tests passing with zero Dart analyzer warnings
- Publication-grade statistical discovery pipeline with Benjamini-Hochberg FDR and Holm-Bonferroni corrections
- Strict chronological data partitioning (50% Discovery, 25% Validation, 25% Frozen Holdout)
- Leave-One-Dimension-Out ablation testing across 8 distinct evidence dimensions
- Zero live-trading safety guardrails guaranteeing purely empirical research and simulation