Horizons
A signal can predict day-30 returns brilliantly while being useless at day-90 — or the reverse. The flat calibration weights collapse all horizons into a single number; this view splits them apart. khalid_index predicts long-term regime, not 30-day P&L. screener_top_pick is the opposite — fast tactical signal, no long-horizon edge. Multi-horizon weighting lets each signal earn its keep at the timeframe where it's actually predictive.
Loading…
Horizon-uplift signals
Signals where the best horizon's weight beats the flat-aggregate weight by ≥15 percentage points. These are the signals that the current flat calibration is mis-pricing — they're predictive at some horizon, just not all of them. Use the recommended horizon for forward weighting.
Multi-horizon matrix
Weight × accuracy × n for every (signal, horizon) pair. Heatmap-tinted by weight: green ≥1.0 (high trust), yellow 0.7–1.0 (decent), red <0.5 (floored / contrarian). Empty cells = no outcomes scored at that horizon yet.
| Signal | Flat weight | day_7 | day_30 | day_60 | day_90 | Best horizon |
|---|---|---|---|---|---|---|
| Loading… | ||||||