research: validate shot context and historical MLB player data

[choigod1023]

Scope

• Research only: no production model changes, cron, raw data publication, merge or deployment.
• Bounded StatsBomb acquisition: 30 matches, 785 shots, 30/30 final scores reconciled.
• Official MLB historical player counts: 2821 player-period splits, four sampled date/count checks passed.
• Exhaustive five-class offline PA diagnostic retains all corrected rows and freezes updates across each date.

Executed results

• Soccer added-context Brier 0.053173 vs provider xG 0.051953: correction not adopted. Conditional shot outcome only, not pregame.
• MLB pitcher-history missingness 994 to 15 of 1177 evaluation PA; batter 161 to 9.
• Seeded league vs seeded batter/pitcher: logloss 1.322904 to 1.307487; Brier 0.684605 to 0.679517. Same 1177 PA, 15 games, two dates.
• Integrated-branch rerun reproduced exact metrics; input database hashes unchanged.
• Small retrospective conditional PA result, NOT evidence of improved game-win/recommendation accuracy or historical point-in-time availability.

Verification

• 782 pytest tests and 60 subtests passed.
• Raw/private SQLite outputs ignored by Git; aggregate Markdown reports and reproduction commands included.
• Independent of unmerged PR218. No production imports or configuration changes.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 78fe610c-1b6e-457f-920d-a766e64a2416

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}