Graded against results

Most golf models ask you to trust the number. This one asks you to check it. Every pre-event projection on MatchWiz is locked before the first tee shot and graded against the real leaderboard after the final round — the same ledger you just scrolled through above, nothing altered after the fact.

What gets frozen before the first tee shot

Before the opening group tees off each week, the model publishes a snapshot: every player's projected rank, win probability, top-finish odds, and make-cut call for that field. Those numbers don't move once the tournament starts. The scorecard can change the leader by eight shots by Thursday afternoon — our pre-event snapshot doesn't budge.

After the final round, automatic grading runs: how many of our projected top-10 players actually finished top 10? Did the make-cut calls land? Where did the winner rank in our pre-event model? The results are posted to the report card — win or lose — with no revisions and no cherry-picked weeks left out. That's the full ledger above.

How to read the finish columns

Each event row shows three hit rates: top-5, top-10, and top-20. They're simple counts. If we projected ten players as our pre-event top-10 picks and seven finished in the real top 10, that event grades at 7/10. The aggregate across all graded events is the season hit rate you see in the summary.

A subtlety worth knowing: golf fields run 120–156 players, so a coin flip on any individual player making the top 10 sits around 8%. The model's top-10 picks aren't random draws — they're the players DataGolf's strokes-gained ratings flag as the best skill fits for that week's field. Whether the hit rate over time clears a random baseline is exactly the question the report card answers.

Make-cut accuracy works the same way: we predict each player as more likely to make the cut than miss it, or vice versa, and grade the call after the cut is made. In a field of 156, roughly half miss the cut — the model has to be better than 50% to clear the baseline.

Group picks — the round-by-round signal

The group picks board tracks a finer lens. Before each round, the model posts a pick for every 2-ball and 3-ball tee-time pairing: the player projected to post the lowest score in that group for that round. The pick locks when the group tees off. It's graded once every player in the group has finished.

Ties are handled dead-heat style. If two players tie for low score in a 3-ball, both are co-winners — a pick that lands in the tie earns 0.5 credit, not 1.0. That's the same settlement most books use for group betting markets, so the credit column reflects what an actual wager would have returned, not a generous rounding-up.

The random baseline for a 3-ball is 33% — one-in-three. For a 2-ball it's 50%. The mix of two-balls and three-balls determines where the average baseline sits for a given week or season. If the model's pick rate is running above that baseline, the skill ratings are doing real work.

Group picks — graded record

Across 1000 graded groups (341 two-balls + 659 three-balls), our picks hit 58% of the time — against a random baseline of around 39%. Dead-heat credit across all picks comes to 52%. On the 485 rounds where we had published book odds, the picks ran at +74.1 units.

The pre-event model — graded record

Season track record

Across 166 graded PGA Tour events, when MatchWiz projected a player inside the top 10 before the tournament, 28% of those picks actually finished top 10. Make-cut accuracy — players modeled as more likely to make the cut than miss it — has landed at 72% across the same sample. Top-5 hit rate sits at 22%. Every projection is frozen before the first tee shot and graded against the official final leaderboard — no revisions, no cherry-picking. Pull up any past event below and read exactly what the model said before a single ball was hit.

What the record doesn't show

The aggregate hit rate hides variance. A 60% top-10 hit rate across twenty events still means eight events where the majority of our top-10 calls missed. Golf is high-variance — strokes-gained skill is the best predictor the sport has, and even the best arms in golf have bad weeks. The model can be right about the likely winner and still watch that player miss the cut.

The honest use of this report card is the same as any graded record: look at the size of the sample, check whether the rates sit above a random baseline, and use that to calibrate how much weight to put on the model's calls any given week. It's a research signal. It's not a guarantee, and we won't frame it as one.

How to use this page as a research tool

Expand any event row to see the full graded board for that tournament: every player the model projected, their pre-event rank, and where they actually finished. You can see whether the miss was the model projecting a player too high (overconfidence on a known quantity) or too low (a sleeper who outplayed their SG). Patterns across events tell you more than any single week.

The group pick rounds show the round-by-round signal at its most granular. A round where the model went 18-for-25 is different from one where it went 12-for-25 — and scrolling the individual picks inside a round shows you exactly which arms the model was highest on and whether they delivered. That's the kind of calibration that takes a week to build from scratch but takes a minute to read here.

For this week's live projections and the current value plays, head to the Golf board. For the methodology behind the numbers, how it works goes through the SG model, the simulation, and the value-play threshold in full.