This model update improves how PropsBot generates Confidence Scores.
Previously, all 59 player prop models used the same default training settings. Now, each market gets its own optimized
settings through an automated search process that tests dozens of configurations and picks the best one for that
specific market.
NHL and NBA models have been retrained and promoted with the new tuning approach.
4 of 8 markets showed improvement, with the best gain of +4.7 percentage points on Shots on Goal:
14 of 16 markets showed improvement, with consistent accuracy gains across the board:
Different markets behave differently — a high-volume market like points has very different characteristics than a rare
event like triple doubles. Using the same training settings for both leaves accuracy on the table. Per-market tuning lets
each model find the configuration that works best for its specific data.
NFL and MLB models will be retrained with per-market tuning ahead of their respective seasons.
