Reading the May 12 CWS Field Through the Model

Welcome to The Margin. Methodology piece, not a picks piece. The CBB model has a clean opinion on D1Baseball's May 12 projected College World Series field, and this post walks through where its number lines up with the seeding and where the two come apart.

Below: the most-likely bracket, a 50,000-sim Monte Carlo, rotation strength by starter slot, and the injury deltas the seeding ignores. Every team and every regional lives on the CWS Odds page; the upstream Field-of-64 projection (which teams make it, who hosts) lives on the Field of 64 page.

Where the bracket sits

The chalk projection has UCLA as the title winner. They're the #1 overall, they host, the sim backs it. UCLA wins the bracket outright in 20.05% of simulated runs, reaches Omaha in 67.3%, and survives the regional in 80.9%. Georgia Tech follows at 14.54%, Georgia at 11.24%. After that, five teams cluster inside a single percentage point: Texas A&M 8.16%, Texas 8.06%, North Carolina 7.72%.

The rating gaps do not match the seeding gaps. UCLA's PEA composite is 7.49, Georgia Tech is 7.04. PEA composite is a per-team rating where one point translates to roughly nine percentage points of single-game win probability. A 0.45-point gap is real, but it's small. Day-to-day starter variance is bigger, typically ±2 rating points depending on who's pitching. UCLA is the cleanest title pick on the board. The gap to Georgia Tech is less than a quarter of one good or bad start.

The price implications follow directly from the simulation. 20.05% prices to +399 in American odds (the model's fair price on a one-shot title bet). Anything tighter is paying less than the model's number. Georgia Tech at 14.54% prices to +588 fair. Texas A&M at 8.16% prices to +1126, Texas at 8.06% to +1140, UNC at 7.72% to +1195. Small-market futures pricing tends to lag rating consensus most in the 5-through-8 seed range, which is where the model has the largest gap to typical book prices.

Where the seeding diverges from the model

The favorite-wins-every-game path is chalk by construction. Variance lives in the 50K Monte Carlo, and three regionals show the largest split between the seed line and the ratings.

Hattiesburg is the regional with the biggest gap. Southern Miss as the 1-seed wins it 47% of the time. Oklahoma as the 2-seed wins it 38%. Neither carries an injury hit. That's a near coin flip wearing a 1-seed designation.

Eugene lands in the same neighborhood. Oregon as the 1-seed wins 50%, Arkansas as the 2-seed wins 37%. Arkansas brings SEC pedigree, Oregon has home field, and if both aces start the regional final, the ratings put it inside a single PEA point.

Tallahassee is the cleanest 2-seed case on the board: Florida State 54%, Southern California 32%. Both rosters are healthy. Tighter than the seed line implies.

The 2-seed case that does not survive scrutiny is Wake Forest at 37% in Gainesville. The Monte Carlo gives it the same surface weight as the other regional upsets, but Wake carries a -0.66 injury delta (roughly a 6pp per-game hit). The injury-adjusted number is materially below the surface 37%. Florida's regional is tighter than the seed line, but the path through Wake is not where the value sits.

Pitching edge: where the chalk hides a problem

Rotation strength by starter slot is where the title contenders separate. Two cases stand out.

Georgia Tech's Friday starter is the contender problem nobody is pricing. T. McKee carries a 4.18 ERA, the worst ace ERA in the top eight. Their depth saves them: #3 J. Blakely sits at 3.18, one of the better #3s in the field. But in a regional final or super regional Game 1, the ace pitches, and Georgia Tech starts McKee against the opponent's best. The 14.5% title number sits on depth, not on the matchup the bracket forces.

Texas is the contender hiding in plain sight. D. Volantis (L) carries a 1.87 ERA, second-best ace in the field and better than UCLA's W. Moss at 1.99. Texas projects to 8.06% on title odds and gets priced like a 6-seed. The rating gap to UCLA does not justify a 12-point title-odds gap when Texas owns the better Friday matchup.

A few rotation notes alongside:

• Florida's A. King is the best Friday starter in the field at 1.73 ERA. Florida projects to 2.34% on the title and the staff depth doesn't support more. King is the reason the Gainesville regional projects tighter than the 10-seed line.
• Auburn carries the most balanced rotation in the top 10. J. Marciano (L) 2.26, A. Alvarez 2.56, A. Petrovic 3.14. No weak link. The 4.74% title number underrates the staff in a double-elimination format.
• North Carolina's #3 F. Boaz sits at 6.64 ERA, the worst depth piece among contenders. A real problem when the bracket forces a fourth game.

Injuries the seeding does not price

The model carries explicit rating-point deltas for pitcher health. One rating point is roughly 9pp of single-game win probability. Four entries matter for the May 12 field, and two are confirmed Tommy John surgeries.

Oregon State: Dax Whitney, season-ending Tommy John. Confirmed by oregonlive, World Baseball Network, and Statesman Journal. The model carries a -0.99 delta. The chalk path keeps Oregon State as the Corvallis host, but the Monte Carlo gives them only 52.8% to reach the super regional and 19.8% to reach Omaha. Those are soft numbers for a 14-seed national-seed host, and Whitney's surgery is the reason.

NC State: Jacob Dudan, also Tommy John. Confirmed by SI and Backing The Pack. The model carries a -0.83 delta. NC State is not in most futures conversations, but two TJ confirmations on the same weekend in the same field is the kind of data point that does not surface in the committee's seed line and does surface in elimination-game variance.

Michigan: Kurt Barr, missing without explanation. A -2.50 delta, the largest in the field, roughly a 22pp per-game hit. Barr leads the rotation by innings and has not appeared in 19 days. No confirmed news, but the absence is the data point. The chalk path has Michigan losing the regional regardless.

Wake Forest: Blake Morningstar, 18 days absent. No confirmed surgery, just a top-5 starter who hasn't pitched. The -0.66 delta is why the 37% surface upset number in Gainesville does not survive the injury adjustment.

UCLA carries a small -0.46 that the 20.05% title number already reflects.

How the model gets built

The CBB model trains on 70,000+ recent D1-to-D1 games and anchors to PEA ratings as they sit on the day of the run. Each game gets scored on the rating gap, adjusted for home field (worth roughly a 6% single-game boost), with starter variance and series fatigue baked in. The 50K Monte Carlo runs that forward through D1Baseball's projected field. The chalk projection drops the random noise to show the favorite-wins-every-game path. Full parameters and training notes live on the methodology page.

Whether the 20.05% on UCLA is trustworthy is a calibration question, not a vibe question. The answer lives on the calibration page: bin charts, posterior updates, and a public ledger of where the model has been wrong.

What's next

The projection gets replaced by the real bracket on Selection Sunday, May 25. Until then, the May 12 read is the cleanest available picture of what the committee should produce — and the places the ratings push back against the seed line. Next post drops the day the field is official: same model, same calibration discipline, real bracket.