Search for a signal or pick from recent runs below.
How to read Feature Reports
Feature reports split your signals into buckets based on relative feature values (feature / sumN_diff) at signal time.
For example, if diff_15min = 2% and sumN_diff = 5%, the relative value is 0.4.
Buckets (B1, B2, B3, B4) divide signals by quantiles of this ratio. B1 = lowest relative values, B4 = highest.
What to look for:
- PnL gradient: If avg_pnl increases from B1→B4 (or decreases), the feature has predictive power.
- Consistency: A bucket that's profitable in most months is a reliable signal.
- Spread: Large PnL difference between best and worst bucket = strong discriminator.
- Actionable insight: If B1 is consistently negative, filtering out low-ratio signals could improve PnL.