SultraxAI Methodology
This page describes how SultraxAI generates BUY/SELL signals, how we measure their outcomes, and how we report win rates. The intent is full transparency on our methods.
Signal generation
The signal model is a rule-based multi-factor system. For each symbol in our universe, on each scanner refresh (every 60 seconds during market hours), the model evaluates:
- Momentum state via RSI and MACD relative to thresholds adjusted per symbol's typical volatility profile.
- Trend regime via ADX (≥ 25 for trending classification) and moving-average alignment (price vs 20-EMA vs 50-EMA).
- Volume confirmation: current period volume vs 20-period rolling average.
- Volatility expansion: ATR vs recent baseline.
- Price-action context: position relative to recent swing highs/lows.
A signal fires when the composite score exceeds a defined threshold. Signal strength (BUY, STRONG BUY, SELL, STRONG SELL) reflects how far above threshold the composite score is.
Signal logging and back-checking
Every signal is timestamped at the moment of generation with the underlying price, all input indicator values, and the signal strength. After generation, the system records the underlying price at three fixed horizons:
- 1 hour after signal
- 4 hours after signal
- 24 hours after signal
A "win" for a BUY signal at horizon X = price at horizon X is higher than the signal price. For a SELL signal at horizon X = price at horizon X is lower than the signal price. No threshold beyond direction. No exclusion of "small" moves. Every signal counts.
Published win rates
The platform-wide win rate at each horizon is calculated as (total wins / total resolved signals) × 100. Per-symbol win rates require a minimum of 30 logged signals before publishing to avoid small-sample noise. Symbols below that threshold are not displayed in the per-symbol table.
Win rates are recomputed daily. The numbers you see on the dashboard reflect all signals ever logged on the platform — no rolling windows that could be used to hide bad periods.
What the win rates don't tell you
Win rate alone is not a profitability metric. A 60% win-rate strategy with negative expectancy (average loss > average win × win rate / loss rate) is not profitable. We publish both win rate and average return per signal so traders can compute expectancy themselves.
Win rate at 1h is more reliable signal direction; at 24h, broader market regimes dominate and the platform's signal contribution is partially diluted. We publish all three horizons so traders can see how the model performs across time scales.
Limitations we acknowledge
- Slippage and fills aren't modeled. Real execution at the signal price requires either fast execution or limit orders. Actual trader P&L will differ from signal-price-based win rates.
- The universe is curated, not exhaustive. We track ~100 US stocks selected for liquidity and retail relevance, plus 12 major cryptos. Less-liquid names aren't covered.
- Regimes change. Past win rates don't guarantee future win rates. We publish full history so traders can see how the model performed across market regimes — but the future could differ.
- Sample sizes vary. Some symbols have hundreds of logged signals; others have tens. Per-symbol numbers below 30 signals are not displayed.