Scoring Methodology
GemStox does not produce a single black-box number. It runs 14 independent strategy theses on the same market and fundamentals data, then lets a consensus — not any one model — decide. Here is exactly how the score is built.
The core idea: many theses, one consensus
Every stock is scored across 14 independent classes. Each class is a pure function of the same shared data bundle (price history, fundamentals, dividends), so no two theses double-count the same signal. The Gem Score is the consensus of the classes that have enough data to vote — the higher the agreement, the higher the conviction.
The 14 theses
- Deep Value. Graham–Dodd–Buffett–Piotroski: P/B, Graham number, earnings yield, FCF yield, Altman Z.
- GARP. Growth at a Reasonable Price: PEG, revenue growth, earnings surprise, forward-P/E compression.
- Momentum. Trend following: 12-month momentum, moving-average alignment, relative strength, volume.
- Mean Reversion. Oversold bounce: RSI, Bollinger %B, distance to the 200-day MA, exhaustion.
- Quality Compounder. Wide-moat: ROE, gross margin, FCF quality, debt vs FCF, growth durability.
- CAN SLIM. O'Neil IBD: current + annual EPS, new highs, supply/demand, relative-strength rating, institutional.
- Magic Formula. Greenblatt: earnings yield + return on capital, ranked across the universe.
- Low-Vol Defensive. Defensive: beta, max drawdown, annualized volatility, Sortino, dividend cushion.
- Catalyst / Event. Event-driven: filings, news, earnings proximity, price and volume catalysts.
- Dividend Growth. Income: yield quality, growth streak, FCF coverage, Chowder, balance-sheet safety.
- S-Curve Growth. Lifecycle: trajectory, position on the curve, persistence, scale, divergence.
- Piotroski F-Score. The 9-point F-Score — a crisp, binary read on fundamental improvement.
- Operating Leverage. Whether revenue growth is converting to expanding operating income.
- Net Payout Yield. Net cash returned to shareholders relative to value (buybacks + dividends).
How a class score becomes 0–100
Within each class, several sub-signals are scored 0–100 and combined with fixed weights. Signals are compared against sector-relative benchmarks (a tech P/B is judged against other tech, a bank's against other banks), so the score is meaningful across the whole universe. If a class lacks the data it needs, it simply doesn't vote — the consensus denominator shrinks accordingly.
From score to signal
The public pages show the consensus Gem Score and the best-fit thesis. The full Day Pass adds the layer most summaries skip: each thesis emits a probability score, the stack computes an optimal position size (Kelly-weighted), and pre-calculated stop and target levels — so a high-conviction score comes with a concrete, sized plan, not just a number.
See it live
Try the free calculators, or open any tracked stock.
Educational explanation of a quantitative model. Nothing here is investment advice. Past performance does not guarantee future results.