How good are these forecasts?
Every forecast scored on seasons the model never saw while its settings were chosen.
Against the betting markets
Each market's price before the start, scored on the same finished games as the model. A negative figure means the market was more accurate.
| Market | Since | Priced | Model log loss | Market log loss | Model vs market |
|---|---|---|---|---|---|
| Polymarket | 10 Apr 2025 | 30 | 3.983 | 3.867 | −3.0% |
| Kalshi | 20 Feb 2025 | 74 | 4.148 | 3.972 | −4.4% |
This is strokes gained against the field from round scores, not shot-level Strokes Gained from ShotLink: it knows how many strokes a player took, not where they came from.
The cut is the top 65 and ties after two rounds; the Tour's other rules (the Tour Championship's starting strokes, no-cut events) are not reproduced.
Rounds on the other men's tours (PGA Tour, DP World Tour, LIV) count toward each player's skill, since the best players split their time; only this tour's events are forecast and scored.
Withdrawals during an event count as missing the rounds they didn't play.
Course fit (a long hitter's course, a putter's course) is not in the model.
Questions and answers
How accurate are PodiumOracle's PGA Tour predictions?
On 137 tournaments from held-back seasons (2024 to 2026), which the model never saw while it was tuned, its favourite won 14% of the time in fields of about 120, and it gave the eventual winner 4% on average. Its log loss was 4.303, against 4.715 for treating every player as equal and 4.427 for recent form. Against the betting markets on the same finished events, its log loss was 3.983 against Polymarket's 3.867 over 30 tournaments and 4.148 against Kalshi's 3.972 over 74 tournaments (lower is better).