Horse racing as an ML ranking problem: 1.18M runners, walk-forward validation and a very strong market baseline [D]

A long-time lurker in a dev community reported on the first live day of rebuilt reports for horse racing as an ML ranking problem. The Top 1 selection achieved a 43.5% strike rate, winning 10 out of 23 races after one non-runner. Furthermore, the winner was found within the Top 1-3 selections in 16 out of 24 races, demonstrating promising results for the machine learning approach.

This report details the first live day results for a machine learning model in horse racing, offering concrete performance metrics unlike theoretical discussions.