This NHL model update improves PropsBot’s player prop Confidence Scores by introducing opposing goalie strength as a core feature across seven skater markets — creating sharper projections and more realistic game environments.
• Added opposing goalie strength features to 7 NHL skater prop models
• Models now differentiate between elite starters and backup goalies
• Removed cross-sport team_id data leakage issue across all models
• player_total_saves retrained with bug fix only — no goalie strength inputs added
Opposing goalie strength has been added to the following NHL skater prop markets:
• Points
• Goals
• Assists
• Shots on Goal
• Blocked Shots
• Power Play Points
• Anytime Goal Scorer
Previous NHL models treated all opposing goalies equally within team defensive context. This update introduces direct goalie-level differentiation — one of the most impactful contextual drivers in NHL scoring environments.
Elite starters suppress shot quality and conversion rates differently than backups. This update allows the model to reflect those differences directly instead of indirectly.
• Retrained 7 NHL skater models with goalie-strength features
• player_total_saves retrained separately (bug fix only)
• Removed cross-sport team_id leakage across all sports models
The team_id leakage fix improves model isolation and eliminates cross-sport data contamination risk. This strengthens signal integrity across the entire modeling system — not just NHL.
Goalie quality materially impacts:
• Expected goal rates
• Shot volume
• Shooting efficiency
• Power play conversion rates
• Correlation across skater markets
When a confirmed backup starts, market pricing often lags behind contextual scoring impact. By integrating goalie strength directly into model features, Confidence Scores now better reflect real scoring environments rather than generalized team defense assumptions.
This increases separation in high-disagreement situations and sharpens projections when lineup confirmations occur.
Removed a data leakage bug where team_id was leaking into training features across all sports. Future
enhancements may include team identification features, but this one is simply a key for joining data and is not a feature.
