Learned Weights
Recursive least squares (RLS) with forgetting factor — per-regime coefficients learned from realized outcomes.
This is the calibration introspection panel. Read-only view of the learned weights per regime. Cold regimes (n_obs < min) don't contribute a learned boost — the ranker falls back to the static composite for those. Weights are bounded to ±2 and the predicted boost is clamped to [0.5, 1.5] regardless of how confident the model becomes.
Loading…