Post-mortems on what the models got wrong, and notes on what changed as a result.
Live graded results, persisted backtest evidence, calibration coverage, and production readiness by sport.
Baseline random-forest home-run metrics, the GPU training plan, and the before/after comparison slot.
