Under the Hood
How the scanner thinks.
From 3,500 tickers to your shortlist in under 4 minutes.
The Three-Phase Funnel.
An intelligent narrowing process. Each phase eliminates noise so the next phase can go deeper.
Phase 1
Universe Screen
3,500+
Every session starts with the full US equities universe: NYSE, NASDAQ, large, mid, and small caps. Phase 1 runs a 5-day bulk prescreen with a 4-minute timeout. Minimum filters: $2 price floor, $50M market cap, 100K average daily volume.
Phase 2
Technical Filter
300-500
60-day technical analysis: SMA crossovers, RSI extremes, ADX trend strength, volume confirmation. This phase removes tickers with no actionable setup. Only technically interesting names advance.
Phase 3
Full 17-Engine Analysis
Top 150
The surviving tickers get scored across all 17 engines. Each engine runs independently. The meta-scorer weighs the ensemble, checks for redundancy, and produces a 0-100 conviction score.
The 17 Engines.
Each engine specializes in a different dimension of market behavior. They score independently. The meta-scorer weighs the ensemble.
Technical
Technical52 indicators including RSI, MACD, Bollinger Bands, Keltner Channels, ATR, ADX, stochastic oscillator, Williams %R, MFI, Donchian channels, and moving average systems across multiple timeframes. The broadest single-engine coverage in the system.
Momentum
QuantitativeMeasures trend strength using directional movement, price acceleration, and relative performance vs sector and broad market benchmarks. Strongest in trending regimes.
Mean Reversion
QuantitativeIdentifies oversold and overbought extremes using statistical z-scores, Bollinger %B, distance from key moving averages, and reversion probability models.
Volatility
TechnicalCompares implied vs realized volatility, detects volatility regime shifts, contraction/expansion cycles, and estimates breakout probability.
Risk
QuantitativeDrawdown analysis, Value at Risk, tail risk metrics, correlation exposure to market factors, and portfolio-level risk contribution estimates.
Sentiment
SentimentAggregates social media sentiment scores, macro sentiment indicators (yield curve slope, VIX level, risk-on/risk-off gauges), and earnings announcement sentiment.
Factor
FundamentalEvaluates exposure to value, growth, quality, and profitability factors using Piotroski F-Score, Altman Z-Score, DuPont decomposition, and earnings quality metrics.
Anomaly Detection
ML32 detection functions using isolation forests and statistical z-score methods to identify outliers across price, volume, fundamental ratios, and cross-sectional data.
Microstructure
TechnicalAnalyzes volume profiles, VWAP behavior relative to price, and identifies signatures of institutional accumulation or distribution.
Intermarket
QuantitativeMaps cross-asset correlations between equities, treasury yields, commodities, currencies, and sector ETFs. Detects when correlations break down or strengthen.
Macro
FundamentalReads yield curve regime, Fed policy direction, recession probability indicators, and sector rotation signals driven by economic cycle positioning.
ML Ensemble
MLXGBoost, LightGBM, and Random Forest models vote on every signal. Trained on 100+ engineered features. Retrained daily with walk-forward validation.
Network Effects
QuantitativeMaps supply chain relationships, sector contagion paths, and correlation clustering to identify when a move in one stock is likely to pull others.
Statistical Arbitrage
QuantitativePairs trading signals using cointegration tests and spread analysis. Identifies when historically correlated stocks diverge beyond statistical norms.
Data Engine
TechnicalMulti-source data aggregation with automatic failover across yfinance, Alpha Vantage, and Polygon.io. Validates data quality before passing to analysis engines.
Meta-Scorer
MLCross-engine signal correlation analysis and redundancy detection. Prevents score inflation when multiple engines fire on the same underlying pattern.
Quant
QuantitativeQuantitative strategy signals combining momentum factor, value factor, and quality screens into systematic long/short frameworks.
Regime Detection.
Markets don't behave the same way in a bull run as they do during a selloff. Hidden Markov Models classify the current regime into one of four states, then adjust engine weights accordingly.
Risk-On / Trending
Momentum and trend-following engines weighted higher. The system leans into directional moves.
Risk-Off / Volatile
Risk and volatility engines weighted higher. Defensive positioning, tighter stops.
Mean-Reverting / Choppy
Mean reversion engines weighted higher. Range-bound strategies, fading extremes.
Low Volatility / Quiet
Factor and fundamental engines weighted higher. Quality and value over momentum.
Engine weights shift automatically when the regime changes. Regime-stratified optimization validates each weight set against historical data for that specific regime, preventing overfitting.
The Learning Loop.
Static scanners run the same rules every day regardless of what the market is doing. Signal Alchemist retrains daily.
Market closes
New price, volume, and fundamental data ingested.
Feature engineering
100+ features extracted: macro indicators, cross-asset correlations, earnings signals, microstructure metrics.
Model retraining
XGBoost, LightGBM, and Transformer sequence models retrain on updated data.
Hyperparameter search
Optuna optimization tests thousands of parameter configurations.
Walk-forward validation
Every model change tested on data the model hasn't seen. Real transaction costs included.
Monte Carlo simulation
Bootstrap resampling generates probability distributions and confidence intervals.
Weight optimization
Regime-stratified optimizer updates how much each engine matters.
Anti-pattern check
Conditions historically associated with drawdowns are flagged.
Live deployment
Updated models serve the next morning's scan.
Validation Methodology.
Walk-Forward Testing
Every backtest uses strictly out-of-sample data. Multi-window evaluation across different market periods catches models that only work in one environment. Historical S&P 500 membership data back to 2015 ensures we're not accidentally testing on companies that didn't exist in the test period.
Monte Carlo Confidence
Bootstrap resampling generates thousands of possible outcomes for each signal. Instead of a single expected return, you see the distribution of possibilities. Kelly criterion position sizing uses the statistical edge to calculate how much capital to allocate.
Survivorship Bias Elimination
Most backtesting platforms only test on companies that still exist today. That silently inflates returns by 1-3% per year. We use historical index composition data to include delisted, acquired, and bankrupt companies in our test universe. The results are honest.
Quality Assurance.
Six autonomous agents monitor system health around the clock.
Data Quality Sentinel
Runs 8 checks during market hours on price freshness, volume anomalies, and fundamental data staleness. Quarantines bad data before it reaches the engines.
Calculation Verifier
130+ unit tests audit every scoring formula daily. Catches drift in calculations, rounding errors, and edge cases.
API Health Monitor
Tracks endpoint response times, error rates, and availability. Alerts on degradation.
Learning Loop Monitor
Daily health assessment of the model retraining pipeline. Verifies convergence, checks for training failures, and validates output distributions.
Parity Checker
Static analysis of frontend and backend code to detect when they fall out of sync. Catches display bugs before users see them.
Performance Regression Detector
Weekly trend analysis on signal accuracy, scan times, and system performance metrics. Flags degradation before it becomes visible.
Data Sources.
Signal Alchemist aggregates data from multiple providers with automatic failover.
Price Data
yfinance (primary), Alpha Vantage (fallback), Polygon.io (fallback). Automatic failover if the primary source is slow or unavailable.
Fundamentals
SEC EDGAR filings, analyst consensus estimates, quarterly and annual financial statements, insider transaction data.
Options
Real-time option chains with full Greeks (delta, gamma, theta, vega, rho), implied volatility surface, max pain, and unusual activity detection.
Sentiment
Social media sentiment scoring across Reddit and financial forums. Macro sentiment via yield curve slope, VIX regime, and risk-on/risk-off indicators.
Macro
FRED economic data (GDP, employment, inflation, rate decisions), sector rotation indicators, and recession probability models.