OkkanaiPai: Building Mahjong Intelligence from 5 Million Real Hands—Without the Cloud

OkkanaiPai demonstrates that game-assistance tools don't require live inference or cloud infrastructure. By distilling patterns from nearly 5 million competitive mahjong hands into an offline, coefficients-based system with 0.83 AUC, it proves that statistical calibration can deliver meaningful real-time guidance that fits into actual gameplay without disruption. The project signals a broader trend: niche communities increasingly building specialized, privacy-preserving tools using their own competitive data.