📊 EQUILYTICS RACE READINESS REPORT

Algorithm: Trained Random Forest — Option 1 (May 2026), exported from Python sklearn
Accuracy: MAE 1.49 ± 0.35 positions (race-grouped 5-fold CV) | 39 features | 400 trees | 116,282 nodes
Training Data: 771 pre-cutover RAPRO rows (658 unique races, Jul 2024 – Jan 2026)
Top Features (impurity %): Margin_won (33.6%) • Speed Rating LS (14.8%) • Q-Rating LS (14.6%) • Place SR (12.7%) • Margin LS (6.0%) • Q×SR (4.3%) • Preparation No. (3.8%)
No-Data Handling: First-starters / long spell returners (no Q-Rating LS, SR LS, or Margin LS) → Not Ranked (matches Python pipeline)
Python Parity: <1e-6 prediction drift, ranks identical on validation races
Sign Convention: Lower predicted position = better expected finish (rank 1 = strongest)