The ScratchOffsNY rank prediction model is validated daily on a held-out next-day snapshot using walk-forward cross-validation. Over the last 4 validation days, the model averaged 92.0% pairwise ranking accuracy with a Kendall tau of 0.839 and an average top-10 overlap of 8.3 of 10.
Pairwise ranking accuracy measures, across every pair of games, how often the model's predicted ordering matched the next-day true ordering. 50% is random; 100% is perfect. Industry research on lottery EV ranking sits in the 55–70% range, so values above 80% indicate that the Smart Score factors are picking up real prize-pool dynamics, not noise.
Kendall τ is the rank correlation between predicted and observed ordering for the entire active game set, in the range −1 to 1. Top-5 / Top-10 overlap counts how many of our predicted top picks landed in the actual next-day top picks. Rank MAE is the mean absolute error in integer rank position.
| Date | Pairwise | Kendall τ | Top-5 | Top-10 | MAE | Displacement |
|---|---|---|---|---|---|---|
| 2026-04-19 | 94.4% | 0.887 | 4/5 | 9/10 | 12.9 | 2.7 |
| 2026-04-18 | 88.8% | 0.777 | 3/5 | 8/10 | 14.2 | 5.5 |
| 2026-04-17 | 92.0% | 0.839 | 4/5 | 8/10 | 15.7 | 3.9 |
| 2026-04-16 | 92.7% | 0.854 | 4/5 | 8/10 | 15.3 | 3.5 |
The production rank model is a depth-2 gradient boosted tree ensemble with the following live hyperparameters:
Inputs are the 27 Smart Score factors (see the methodology). Targets are the next-day Smart Score recomputed from updated NY Lottery prize-pool data.
This dashboard is generated server-side from ml/rank_prediction_model.json and refreshed every cycle. Researchers and journalists may cite the numbers under CC BY 4.0; please link back to scratchoffsny.com/methodology/performance.
Data sourced from nylottery.ny.gov and NY Open Data. Independent, non-commercial project. Play responsibly — NY Problem Gambling · 1-877-8-HOPENY.