Aronimink Golf Club, final round of the PGA Championship. Alex Smalley leads at -6, with a six-deep cluster at -4 that includes Jon Rahm, Ludvig Aberg, Aaron Rai, Nick Taylor, Matti Schmid, and Smalley himself one stroke clear. Rory McIlroy and Xander Schauffele sit T7 at -3. Scottie Scheffler is buried at T23, six back at -1.
That last sentence is the whole story. The board says Scheffler is +350 to win. Our model says that price is roughly three times too short.
The PGA tool at golf.propsbot.ai updates every few hours through tournament week. It is free, no signup, no paywall, and runs as a separate product from the main PropsBot.AI daily-pick platform. This week it has the entire Aronimink field modeled with Monte Carlo simulation, Bayesian course-fit priors, and strokes-gained projections. Here is what the model is actually saying about the closing 18 at the year’s second major.
The Headline: The Model Is Fading the Chalk
Going into Sunday at a major, books typically widen prices on the leaders to discourage late action and compress prices on the contenders. Our model is telling us they have compressed too far. Of the players priced inside +2000 on FanDuel, every single outright line currently flags as overpriced, with five of them tripping both the z-score and IQR anomaly detectors. Here is the leaderboard the model actually generated, sorted by mispricing.
| Player | Market | Model Win Prob | FanDuel Line | Implied Prob | Edge | Direction |
|---|---|---|---|---|---|---|
| Scottie Scheffler | Outright | 8.53% | +350 | 22.22% | -13.69% | FADE |
| Rory McIlroy | Outright | 3.16% | +650 | 13.33% | -10.17% | FADE |
| Jon Rahm | Outright | 3.35% | +1200 | 7.69% | -4.34% | FADE |
| Ludvig Aberg | Outright | 2.07% | +2000 | 4.76% | -2.69% | FADE |
| Cameron Smith | Top 20 / Make Cut combo | 15.27% (Top 10) | +6500 outright | 1.52% | positive | VALUE |
Why Each Edge Exists
Scottie Scheffler (+350 FD): FADE
This is the largest mispricing on the entire board, and the only player on the chalk side with a positive model confidence score (84.3) despite being six back. The model loves Scheffler’s Aronimink course fit. The learned-fit number is 99 out of 100, his strokes-gained approach (1.18) leads the projected field, and the Monte Carlo simulation gives him an 8.5% chance to run the table. The problem: at +350, the book is implying 22.2%. To get there from six back at a major, you essentially need to play a 64 while every player at -4 plays 71 or worse. That happens, but it does not happen 22.2% of the time. Our model says fair is closer to +1000.
Rory McIlroy (+650 FD): FADE
Rory is T7 at -3, three back, and the model still has the outright priced 10 percentage points too short. The 65 in round three was electric, but the path from three back at a major against Smalley (who is playing the best golf of his career this week) and the entire -4 cluster requires more than the simulation will give Rory at this price. The model is much more favorable on the top-10 line at 33.65%, which is where the McIlroy belief shows up.
Jon Rahm (+1200 FD): FADE
Rahm is in the cluster at -4 and has the highest top-5 probability in the model at 25.57%, with a top-10 probability of 45.7%. Those are real numbers and reflect that the model likes him to be around the lead on Sunday. But +1200 implies a 7.69% chance to actually win, and the simulation only gives him 3.35%. The conversion from being in the mix to winning the tournament at a major is brutal, and the book is overpaying for Rahm’s name and his strong R3.
Ludvig Aberg (+2000 FD): FADE
Same logic, smaller edge. Aberg is at -4 with a 24% top-5 and 44% top-10 probability. The outright line at +2000 (4.76% implied) versus his 2.07% modeled win probability gives a -2.69% edge. Worth noting because the IQR anomaly detector still flagged it, but if you want Aberg this Sunday, the top-5 and top-10 markets are where the model would let you express that view.
Cameron Smith: The One Positive Edge
The only player in the entire field this morning who shows up as underpriced on a flagged anomaly is Cameron Smith. He is T11 at -2, six back, with a model top-10 probability of 15.27% and a top-20 of 33%. The flagged market is a combination top-20 plus made-cut prop where the implied book number sits below the model number. His +6500 outright price is roughly fair (model has him at 1.47% win, book at 1.52%). The point is not that Smith is winning the PGA Championship. The point is that on a board where every other priced player flags as overpriced, the model isolated exactly one positive-edge player. That kind of selectivity is what we want from a Sunday final-round model.
How the Model Builds These Numbers
Three components drive every projection on the PGA tool. First, a 10,000-iteration Monte Carlo simulation runs every player’s remaining rounds with stochastic scoring based on their projected mean and standard deviation. Second, a Bayesian course-fit model adjusts each player’s baseline using historical results at Aronimink and similar tracks. For players with no Aronimink data, the prior carries most of the weight; for Scheffler, who has one prior result here, the learned fit (99) plus the prior (70) blend to a posterior of 99 with weight 0.33. Third, strokes-gained projections (off the tee, approach, around the green, putting) feed the expected scoring distributions. Full breakdown lives at the tool methodology page.
Honest Caveat on Sample Size
The golf tool is in free beta and the backtest is small. The current backtest report covers 936 graded matchup pairs across 12 weeks of PGA Tour events, with a Brier score of 0.23 against a baseline of 0.249. That is statistically better than the baseline, but on a 12-week window the confidence interval around that delta is wide. We are not claiming a closing-line-validated golf ROI yet, and we will not until the sample is large enough to support the number. The main PropsBot.AI platform has 218,000+ audited picks across NBA, NFL, MLB, NHL, and PGA. The methodology that produces those track records is the same one feeding the golf model, but it takes time to accumulate the graded volume to publish a real ROI on a low-cadence sport. We will publish it when we have it.
How We Will Grade This Week
By Sunday night, the four fade calls will have resolved. If Scheffler wins from six back, the model is wrong on that pick. If McIlroy wins from three back, the model is wrong on that pick. We will record both outcomes in the public backtest dataset regardless. Our performance methodology page describes how every pick is graded against closing lines and tracked on a continuous basis.
Why This Coverage Is Free
The PGA tool sits outside the paid PropsBot.AI subscription. It is a beta product, and we use it to expand the projection engine to a fifth major sport. The main competitive AI prop tool, Rithmm, does not cover golf at all: no outrights, no matchups, no finishing-position props. That is a meaningful gap during major weeks, and one of the reasons we built ours out as a free standalone. The full Aronimink board (outright, top-5, top-10, top-20, 2-ball and 3-ball matchups, and per-round props) is live now at golf.propsbot.ai, no signup required. For year-round PGA betting context, our main-domain golf prop bets guide covers the broader market structure.
The Final Round Read
Final-round outright lines at a major are not where a model usually finds gold. The books have had four days to calibrate, and the field has narrowed to a few legitimate winners. That is precisely the regime where our model is currently saying: pass on the top of the board, take a position in the top-10 and top-20 markets if you have a view, and recognize that the only positive flagged edge is on a player six back from the lead. Sometimes the best read a model delivers is that nothing here is priced right. This is one of those Sundays.
See the live leaderboard, projected finishes, and every market by player at golf.propsbot.ai. No signup, no paywall, updated through Sunday’s final group.
