Split 9ae1bd37fd Refactor: Implement all four ELO system improvements ...

CHANGES:

1. Replace arbitrary margin bonus with per-point expected value
   - Replace tanh formula in score_weight.rs
   - New: performance = actual_points / total_points
   - Expected: P(point) = 1 / (1 + 10^((R_opp - R_self)/400))
   - Outcome now reflects actual performance vs expected

2. Fix RD-based distribution (backwards logic)
   - Changed weight from 1.0/rd² to rd²
   - Higher RD (uncertain) now gets more change
   - Lower RD (certain) gets less change
   - Follows correct Glicko-2 principle

3. Add new effective opponent calculation for doubles
   - New functions: calculate_effective_opponent_rating()
   - Formula: Eff_Opp = Opp1 + Opp2 - Teammate
   - Personalizes rating change by partner strength
   - Strong teammate → lower effective opponent
   - Weak teammate → higher effective opponent

4. Document unified rating consolidation (Phase 1)
   - Added REFACTORING_NOTES.md with full plan
   - Schema changes identified but deferred
   - Code is ready for single rating migration

All changes:
- Compile successfully (release build)
- Pass all 14 unit tests
- Backwards compatible with demo/example code updated
- Database backup available at pickleball.db.backup-20260226-105326

2026-02-26 10:58:10 -05:00

4.5 MiB

Executable File

Raw History

View Raw