A transparent walkthrough of how every pick is scored — from raw data to final ranking.
The Phoenix model doesn't attempt to predict price movements. Instead it identifies candidates that share strong statistical similarity to a benchmark cohort of stocks that went on to produce 60%+ gains.
This is done using Mahalanobis distance — a multivariate metric that measures how similar a candidate's feature vector is to the distribution of historical winners, accounting for correlations between features.
Candidates closest to the winner distribution receive the highest composite scores and are selected for publication. Lower distance = higher score.
The full screening pipeline runs overnight using end-of-day data from Polygon.io.
OHLCV data pulled from Polygon.io flat files (S3) with REST API fallback. Covers all US-listed tickers. Data is deduplicated and cleaned with incremental updates.
Hard filters: price $0.10–$5.00, float <25M shares, 10-day average volume >500,000, price >10% of 52-week high.
Five features: 5-day momentum, volume surge, float turnover, RSI-14, range ratio. All winsorized at 1st/99th percentile and log-transformed.
Distance computed against benchmark covariance matrix estimated using Ledoit-Wolf shrinkage.
Top 5 by composite score published before 08:00 ET with entry price range, TP target (+60%), SL level (-20%), and model notes.
Close position when price reaches +60% above the published entry price. Tracked to the daily high.
Close position when price drops -20% below entry. Protects against sharp reversal on failed setups.
If neither TP nor SL is hit within 5 trading days, position closes at Friday's closing price.