PGA Performance
Windowed live grades by market, plus persisted model evaluation evidence.
Tournament board — win/top-10/top-20
Open market| Event | Model | Players | Top 10 | Top 20 | Winner actual / expected |
|---|---|---|---|---|---|
| tour_championship_2026 | pga-baseline-v0 | 120 | 33.3% | 66.7% | 4 / 4.00 |
| bmw_championship_2026 | pga-baseline-v0 | 100 | 22.0% | 42.0% | 2 / 2.00 |
| fedex_st_jude_championship_2026 | pga-baseline-v0 | 69 | 15.9% | 33.3% | 1 / 1.00 |
| rocket_classic_2026 | pga-baseline-v0 | 146 | 9.6% | 17.1% | 1 / 1.00 |
| 3m_open_2026 | pga-baseline-v0 | 144 | 8.3% | 16.0% | 1 / 1.00 |
| the_open_2026 | pga-baseline-v0 | 421 | 7.6% | 16.2% | 3 / 2.17 |
Persisted backtests
Evaluation runs stored in Supabase, separate from live grading.
Evaluation runs
| Evaluation | Model | Test range | Accuracy | AUC | Brier / log loss | ROI | Status |
|---|---|---|---|---|---|---|---|
| performance_history_pga_test_gte_2025_outright_top_placement | v2 artifacts evaluated on refreshed store | n/a–n/a | n/a | n/a | n/a / n/a | n/a | candidate |
Accuracy over time