Yankees vs. Rays: The Model Likes the Team in Second Place

Start with the standings, because they point the other way.

Through games of August 24, Tampa Bay is 78-53 and New York is 74-56, three and a half games back and holding a wild-card place. Simulate the remaining schedule twenty thousand times and both clubs reach October in essentially all of them. Tampa Bay wins the division in 73% of those seasons, and when it does it is the American League's top seed 97% of the time.

They meet in 41% of the simulations. Most of those matchups come early: 71% are in the Division Series, 22% in the Championship Series and 6% in the Wild Card round.

They meet in 39% of the simulations in which New York wins the East and 41% of those in which Tampa Bay wins it. What the division race changes is the bye and the ballpark. Tampa Bay holds home field in 73% of the meetings.

With Tampa Bay hosting, here is the answer in all three formats:

New York's advantage grows in the longer formats because its rotation edge appears more often. Tampa Bay's best chance is the three-game format, with every game at Tropicana Field and Fried starting only once.

Two scales appear in this article. The raw figure is the simulation's own output. The calibrated figure adjusts each individual game probability using a monotone map fit on 2,421 game-level forecasts from the 2025 regular season, March 27 through September 28. The map compares those forecasts with the games' actual results. It was fit once, in July 2026, and is not rebuilt on a schedule. Series probabilities are then calculated from the adjusted game probabilities.

Series levels are shown calibrated. Every probability of a scenario change below is reported raw, for reasons the methodology note explains.

What moves the series

We tested four changes to the five-game series, each priced separately. Ordered by size, on the raw scale, with 95% intervals:

Judge is the largest single variable, at 6.0 points. Home field and the loss of both left-handers come next, at 3.7 and 3.5, and they run in opposite directions. The platoon correction is the smallest, at 0.8 points.

The Judge scenario also carries uncertainty the model cannot resolve. It can estimate his value if he returns. It cannot determine when he returns, or how closely his performance will resemble his pre-injury level.

One assumption is built into the base series results rather than listed among them. Relievers who throw more than an inning sit the next day, nobody works three days running, and the off days in each format are built in. That is how these series are simulated, not a scenario applied to them. The 2026 Division Series is scheduled for October 3, 5, 7, 8 and 10, so only Games 3 and 4 fall on consecutive days. That is the one place the rule can bind, and it accounts for 18.3% of the games played. It is also the only game at which any arm is unavailable: at Game 4 the rule sits an average of 1.16 New York relievers and 0.78 Tampa Bay relievers, and at all four other games it removes none. Across the series it costs New York about three tenths of a point, measured at 0.29 with a 95% interval of 0.13 to 0.45 across six seed bases. On this calendar, the resulting effect is small.

Where 57% comes from

Tampa Bay is favored in two of the five games, and they are the same matchup: Drew Rasmussen at home.

Tampa Bay's path is to win both Rasmussen starts at Tropicana Field and steal one of the middle three. New York's clearest advantage is Games 3 and 4, at 67.7% and 60.2%. Mean series length is 4.12 games, which reconciles exactly as the three games always played plus 0.754 plus 0.368.

The rotations

Here is every starter in the series on the quality measure the simulation prices them with. Lower is better.

Set against that 0.331 median, the shape of the matchup is plain. All four New York starters rate better than the median starter. Only two of Tampa Bay's do.

Drew Rasmussen is the exception and he is genuinely good, second-best in the series and better than three of New York's four, and the format gives him two of the five starts. But Peralta and Martinez are the two worst arms on the board by a wide margin, and they start Games 3 and 4 against Fried and Weathers. That is where New York creates most of its advantage.

Two caveats on New York's four. The base case assumes Max Fried and Ryan Weathers are both available to start in October. Both are on the 15-day injured list, Fried with a left elbow bone bruise and Weathers with a left forearm flexor strain. Fried is the nearer of the two: he threw four innings of live batting practice on August 24, and manager Aaron Boone said the next day that he could pitch in the August 29 doubleheader against Boston. Weathers is a further question. The club expects him back in late September and has said he may return as a reliever rather than as a starter, so the base case here makes a specific assumption about him, that he returns and returns to the rotation. If he pitches in relief instead, the Game 4 start goes to someone else and the table above overstates New York. That is one reason the analysis also prices the series without both arms. Removing them costs New York 3.5 points.

And Freddy Peralta's rating reflects his 2026 performance rather than his previous reputation. Traded from the Mets at the August deadline, he has been allowing home runs to left-handed batters at 1.6 times the rate of a typical Tampa Bay starter while walking right-handers at nearly twice.

Does New York's rotation last longer?

No. The deepest starter is Nick Martinez, the lowest-rated starter in the series. New York's four sit within a third of an inning of each other, from 5.22 to 5.55. Tampa Bay's starters span more than a full inning.

Depth and quality point in opposite directions for the Rays, and New York's advantage is quality across a bunched rotation rather than the depth story it would be easy to assume.

The bullpens are not doing the same job

Across the series Tampa Bay's relievers throw about 3.7 innings a night to New York's 3.4.

Most of Tampa Bay's relief innings go to one man. Ian Seymour absorbs 1.7 per game, more than any pitcher on either side. He starts sometimes and relieves sometimes, and that hybrid usage makes him the bridge between Tampa Bay's rotation and its bullpen.

The rest rule reaches him hardest. Because his outings routinely run past an inning, an appearance in Game 3 costs him Game 4, which is the only pair of consecutive days on the Division Series calendar. Tampa Bay leans most on the arm the rule constrains most, and the innings it takes from him move to someone worse.

The lineups

Expected value per plate appearance, by opposing starter:

New York projects lowest against Rasmussen at .271 and highest against Peralta and Martinez at .306 and .304. Tampa Bay projects lowest against Fried at .259; its other three matchup values range from .275 to .288.

Max Fried allows the lowest opposing expected value in the series, at .259. That is a different measurement from the rotation ratings above, which place Cam Schlittler first, and the two do not check each other: each pitcher's figure here is computed against a different lineup and a different mix of platoon matchups, while the ratings are computed against a common league baseline. Weathers at .288 allows the highest figure of the four New York starters. Note that the injury scenario priced earlier removes Fried and Weathers together, so it does not isolate what losing Weathers alone would cost.

The sharpest single matchup on the board is Paul Goldschmidt at .380 against Shane McClanahan, 93 points above New York's .287 lineup average for that game.

The next-best New York bat against him is Cody Bellinger at .352, twenty-eight points back. In the lineup the Yankees would field today, Goldschmidt is the best answer to McClanahan, though with Bellinger that close he is not the only one.

New York uses him primarily as a platoon bat, starting him against every left-handed starter but only eight of twenty-five right-handers, so the lineups priced here carry him only against McClanahan.

Reordering the rotation does not move the problem for Tampa Bay, because Goldschmidt starts against the left-hander whenever that start comes.

Even so, New York does not create most of its advantage against McClanahan. Its highest projected EV/PA comes against Peralta and Martinez, at .306 and .304, in Games 3 and 4.

One caveat runs the other way. The simulation does not make in-game substitutions for hitters, so the same nine batters remain in the lineup after McClanahan exits. A real manager would counter a stacked platoon lineup by changing who bats, not only who pitches.

The automated ranking also omitted Chandler Simpson and Cedric Mullins, Tampa Bay's two most-used position players. That outfield ranking weights home runs at twice everything else, which works against a speed-and-contact hitter like Simpson. Goldschmidt, Simpson and Mullins are all in the lineups used here, because these nines were built from each club's actual recent usage rather than from that ranking. Worth noting anyway: on Tampa Bay's most-used position player, the model's own valuation and the manager's judgment disagree sharply.

And then Judge

Aaron Judge has not appeared in a major-league game since May 31, with a stress fracture in his right rib. On August 3 the general manager would not rule out losing him for the season. On August 5 he was cleared for light activity. On August 18 he was cleared to begin a hitting progression. On August 25 he took swings against overhand pitching in the indoor cage for the first time, having been limited to tee-and-toss work before that. He has not hit on the field, and the Yankees have not set a return date.

If he plays, New York's five-game series probability rises 6.0 points.

The size of that lever is mostly a fact about the man he replaces. Judge takes right field from Spencer Jones, who has 178 plate appearances this season. He is not upgrading an established regular; he is replacing a much less established hitter in a lineup already missing its best bat.

Two things the model cannot do here. It has no concept of rust, so it prices Judge at what he was rather than at what nearly three months without game action might leave. And it cannot tell you whether he plays at all.

Judge is better treated as a range than as a single number. At half of his modeled value, Judge adds about 3.0 points, compared with 3.7 for home field. Across six paired runs, home field remained larger by 0.7 points, with a 95% interval of 0.3 to 1.1. This is a scenario assumption, not a forecast of how Judge would actually perform after returning. The realistic effect could fall anywhere from zero to about six points, depending on whether he returns and how much of his pre-injury value remains.

What the model cannot see