Series Focus · 2026-10-01

How Chicago Beats Cleveland

The White Sox went to Houston and won twice, 6–3 and 7–3: their first postseason series win since 2005. Next is Cleveland, the club that took the Central from them by a single game. The model makes the Guardians a 56–44 favourite. This is the full data-science preview, and its spine is the question that matters in October: where, exactly, can Chicago hurt them? Every pitch Cleveland leaks damage on, every location, every platoon split, the soft middle of their bullpen, a running game the numbers say is wide open, and a game-by-game plan that ties it together. Twenty-seven charts.

Hero graphic split diagonally between White Sox black and Guardians navy: White Sox 44 percent, Guardians 56 percent to win the AL Division Series with the projected starters, and five game chips: Game 1 Martin v Messick 41 percent, Game 2 Kay v Williams 45 percent, Game 3 Burke v Bibee 56 percent, Game 4 Smith v Cantillo 51 percent, Game 5 Martin v Messick 41 percent.

Twenty-one years.

That is how long it had been since the White Sox last won a postseason series, and the bookend writes itself. In October 2005 they swept the Houston Astros to win the World Series. Since then: a Division Series loss in 2008, a Wild Card Series loss in 2020, a Division Series loss in 2021. This week they went to Houston and swept the Astros again, 6–3 on Tuesday and 7–3 on Wednesday. That is a historical note, not a forecasting input, and nothing below leans on it. But it is the right frame for a club this model had rated 25th in baseball in June.

The two wins were built the way short series usually are: on the bullpen. Hagen Smith opened Game 1 with three scoreless innings and Grant Taylor struck out five in a 2⅓-inning bridge. In Game 2 Sean Burke struck out eight in four innings and four relievers covered the last five, Erick Fedde and Jordan Hicks among them with three strikeouts apiece. Neither Houston starter got out of the first. Now the bracket sends Chicago to Cleveland, the second seed, rested after a bye, for a best-of-five that opens on Saturday.

Tale of the tape

Regular season: Guardians 85–77 (run differential +11, won the Central); White Sox 84–78 (+56, one game back)
One-run games: Guardians 27–22, White Sox 25–22
Head to head: White Sox 7–6 in 13 meetings; runs 58–57
Model rating: Guardians 12th of 30, White Sox 17th
2026 form (schedule-adjusted run differential per game): White Sox +0.27 (11th), Guardians −0.03 (16th)
Series odds: Guardians 56%, White Sox 44% with the projected starters (47% team against team)
Bullpen pitches in the week before Game 1: White Sox top eight 303, Guardians top eight 16

I. The number, and how it breaks down

Start with the series, because everything else decomposes it. The model’s own bracket simulation, run after Game 1 in Houston, sent the White Sox to the Division Series 76% of the time and to the Championship Series 36% of the time. Divide one by the other and you have the model’s probability that Chicago beats Cleveland once the matchup is set: 47.2%. That is team against team, at the right parks, in the real 2–2–1 format, before anyone names a starter. Add the projected starters and it falls to 44%, because Cleveland’s Game 1 starter is the best pitcher in the series and he can pitch twice.

Bar chart of how the series ends: White Sox in 3, 10.4 percent; in 4, 18.1 percent; in 5, 15.6 percent. Guardians in 5, 22.1 percent (outlined as the single most likely result); in 4, 19.6 percent; in 3, 14.2 percent. — Every way the series can end. The single most likely result is Cleveland in five; Chicago’s most likely winning route is in four, which means taking both games at home.

The tree underneath that is worth a look on its own. Chicago’s route to the ALCS is not three wins from anywhere; it runs through particular states. Lose Game 1 and the Sox are 0–1 59% of the time, with a 29% series chance. Steal Game 1 and they are at 66%.

Series path tree from 0-0 through every win-loss state to the six terminal results, with node size proportional to the probability of reaching that state and edge thickness to the traffic along it. 1-1 is reached 49 percent of the time, 2-2 38 percent; terminal results: 3-0 10 percent, 3-1 18 percent, 3-2 16 percent, 0-3 14 percent, 1-3 20 percent, 2-3 22 percent. — Every road through the best-of-five. Half of all simulated series pass through 1–1, and 38% reach a fifth game: this projects as a long series more often than a short one.

Dot-and-interval chart of the White Sox win probability per game with the projected starters: Game 1 41 percent, Game 2 45, Game 3 56, Game 4 51, Game 5 41, each with an 80 percent posterior interval roughly 30 points wide; orange ticks mark the team-only probability of 47 percent in Cleveland and 51 in Chicago. — Game by game. Without starters this is a coin flip with a home lean (the orange ticks: 47% in Cleveland, 51% in Chicago). With them it becomes three different games: Messick drags Games 1 and 5 to 41%, and Burke against Bibee lifts Game 3 to 56%, Chicago’s best game of the series.

Which game matters most? Set each one to a win and then a loss with the rest as projected, and the swing in series probability is almost identical across the first four games, 37 to 38 points each. In a short series every game is the big one; the difference is that Game 4 is only played three times in four and Game 5 a little over a third of the time.

Dumbbell chart of Chicago's series probability if it wins or loses each game: Game 1 66 percent v 29 percent; Game 2 65 v 27; Game 3 60 v 24; Game 4 63 v 25; Game 5 66 v 29. Each game's swing is 37 to 38 points. The current 44 percent is marked. Game 4 is played 75 percent of the time and Game 5 38 percent. — The leverage of each game: Chicago’s series chance after a win (black) or a loss (navy) in that game, everything else unchanged. Win Game 1 in Cleveland and the Sox become favourites.

Forty-four per cent is not a verdict, and the model is the first to say so. Redraw the team ratings and the starters’ effects from their posteriors and recompute the series each time, and the White Sox’ chance spreads across most of the range.

Histogram of the White Sox series probability across 20,000 posterior draws, coloured navy below 50 percent and black above; median 43 percent, 10th percentile 18 percent, 90th percentile 73 percent. — The series number across 20,000 draws: median 43%, 80% interval 18–73%. Read this as an outer bound. It treats each published rating interval as independent, because the full posterior’s correlations aren’t stored. What it shows honestly is that the gap between these clubs is smaller than the uncertainty about either.

What has to go right

So what moves it? Hold one rating at its 10th or 90th percentile while everything else varies as before, and the answer is blunt: the clubs, not the starters. If Chicago’s offence is really at the top of its range, the series is 58%; at the bottom, 32%. Run prevention on either side matters about as much. Of the starters, Davis Martin carries the most weight (40–48% across his range), more than Messick himself, because Martin is the most uncertain arm in the series and he pitches twice.

Tornado chart of Chicago's series probability when each factor is held at its 10th or 90th percentile. Team ratings dominate, each spanning about 30 to 60 percent. Starters span a few points each: Davis Martin 40 to 48, Parker Messick 41 to 48, the others 42 to 46. Rotation choices in orange: Fedde instead of Smith in Game 4 46 percent; Cleveland starting Williams in Game 5 46; Burke moved to Game 4 44; Kay in Games 1 and 5 44. — The swing factors, quantified. Grey bars hold one rating at its 10th or 90th percentile; orange bars change one rotation decision. The levers Chicago controls are worth one or two points each. The biggest is letting Erick Fedde take Game 4 rather than an opener (55% for the game, 46% for the series).

II. Two average clubs, one of them hotter

The model rating is recency-weighted across three seasons, so it still carries Cleveland’s recent division titles and Chicago’s 2024 and 2025. This column has tested that weighting on held-out seasons and it predicts better than anything shorter, so the rating stays. But 2026 on its own tells a different story. Put the twelve October clubs on one chart and most lie on a diagonal where rating and form agree. Chicago sits furthest above it of all twelve, +0.32 runs a game better than its rating predicts. Cleveland sits 0.13 below.

Line chart of rolling 20-game run differential per game across the 2026 season for both clubs. The White Sox climb from minus 2.5 early to a peak of plus 2.75 around game 98, drop after the Castillo trade, and finish at plus 1.0 including the two Wild Card wins (orange). Cleveland sits below zero through midsummer, peaks at plus 1.55 and also finishes at plus 1.0. — The season in one line each. Chicago’s peak (+2.75 runs a game over 20 games, peaking on July 20) is the best stretch either club produced; both arrive in October at +1.0.

The parks don’t tilt it. Measured as home-game runs over road-game runs, Progressive Field averages 1.00 across 2022–2026 and Rate Field 0.99. Both are run-neutral over five seasons, though either can swing ten points in a single year. This season Rate Field played a little friendlier to home runs (1.03) than Progressive Field (0.96). That is one season of home-run counts, so treat it as a lean rather than a fact.

Left: run park factors by season 2022 to 2026 for Progressive Field and Rate Field, ranging from 0.89 to 1.14, with five-year means of 1.00 and 0.99. Right: 2026 home-run factors, Rate Field 1.03 and Progressive Field 0.96. — Two neutral parks. Season-to-season noise (Rate Field 1.14 in 2022, 0.93 two years later) is bigger than any lasting difference between them.

III. The arms, pitch by pitch

The rotations are projections, not announcements. Cleveland comes off a bye with every starter rested; Joey Cantillo has spent September as a four-to-six-inning bulk arm. Chicago used Smith and Burke to start in Houston with Anthony Kay in relief, which lines up Davis Martin for Game 1. Martin is on a week’s rest after a 25-pitch outing on the season’s final weekend.

Dot-and-interval chart of the model's starter quality for the projected Game 1 to 4 starters with 80 percent intervals and ranks among 833 pitchers: Messick 0.13 (13th), Burke 0.08 (35th), Williams 0.04 (71st), Cantillo 0.01 (123rd), Smith 0.01 (127th), Martin 0.00 (658th), Kay 0.00 (686th), Bibee minus 0.03 (754th). — The rotations as the model rates them. Messick (13th of 833) is the only starter on either side whose whole interval sits above average. Burke is 35th. Martin, Kay and Bibee are the three the model would least like to see.

Now take them apart. Movement first: horizontal break (arm side positive) against induced vertical break, every pitch of the season.

Eight small-multiple movement plots, one per projected starter, showing every 2026 pitch by horizontal and induced vertical break with pitch-type averages. Messick's four-seamer rides at 16.5 inches with his changeup at 1.6 inches of rise and 15 inches of arm-side fade. Williams pairs a 98 mph four-seamer with a sweeper breaking 11 inches glove side. Burke's four-seamer has 18.6 inches of ride. Smith throws three pitches: four-seam, slider, cutter. — What the ball does. Messick’s weapon is separation: a four-seamer with 16.5 inches of ride and a changeup that drops 15 inches below it while fading 15 inches arm side. Burke’s four-seamer has the most ride in the series (18.6 inches). Williams’s sweeper moves 11 inches the other way from a 98-mph fastball.

Small multiples of stacked bars showing each projected starter's pitch mix by count state: first pitch, ahead, even, behind, two strikes. Messick goes four-seam and changeup with two strikes (44 and 37 percent). Williams leans on the sweeper with two strikes (37 percent). Cantillo goes to his curveball (40 percent). Burke throws his four-seamer 49 percent of the time with two strikes; Smith splits slider 44 and four-seam 41. — What they throw, and when. With two strikes Messick is four-seam/changeup (81% combined), Cantillo goes to the curveball 40% of the time, and Burke leans on the elevated four-seamer for nearly half his two-strike pitches. Predictability is information: a hitter who knows the putaway pitch can sit on it.

Scatter of every pitch type in both rotations: whiffs per swing against xwOBA saved relative to the league on that pitch type, bubble size by usage. Top of the chart: Messick four-seam and changeup, Burke four-seam and Smith slider, each about 60 to 67 points better than league. Bottom: Kay cutter 83 points worse, Messick slider and curveball, Martin changeup and Kay sweeper about 50 to 60 points worse. — Every pitch in both rotations, graded against the league on the same pitch type. Messick’s two main pitches are elite, but his secondaries are not; Kay’s cutter is the worst pitch either side throws regularly.

Four location density plots from the catcher's view with the rulebook zone: Messick's four-seamer centred in and above the middle of the zone (56 percent in zone, 42 percent above three feet); Williams's sweeper running down and away off the zone (41 percent in zone); Burke's four-seamer high in the zone (52 percent in zone, 45 percent above three feet); Smith's slider low and off the plate (37 percent in zone, 142 pitches). — Where the key pitches live. Messick and Burke both pitch up with the four-seamer, 42% and 45% of them above three feet. Williams and Smith live below and off the zone with their breaking balls, which is how chases get made (Smith’s map rests on just 142 pitches).

The other half of every matchup is the hitter. Here are all eighteen projected regulars against every pitch family from each hand, shrunk toward the league so small samples can’t shout.

Two heatmaps of each projected hitter's shrunk xwOBA relative to the league by pitcher hand and pitch family, with plate appearances. White Sox highlights: Miguel Vargas plus 73 against right-handed breaking balls and plus 55 against left-handed fastballs; Munetaka Murakami plus 66 against left-handed fastballs; Colson Montgomery minus 44 against right-handed fastballs and breaking balls. Guardians highlights: Jo Adell plus 41 to plus 47 against everything left-handed; José Ramírez plus 52 against left-handed breaking balls; Travis Bazzana minus 39 against left-handed breaking balls; Angel Martínez minus 45 against right-handed offspeed. — Every hitter against every pitch family. Miguel Vargas is the Sox’s most complete threat (+73 points against right-handed breaking balls on 152 plate appearances). Murakami punishes left-handed fastballs. For Cleveland, Jo Adell crushes left-handed pitching of every kind, and Travis Bazzana is the clearest hole.

Combine the two halves, weighting each starter’s results on every pitch family by how often he throws it to each side of the plate, and you get every starter against the lineup he will actually face.

Dot-and-interval chart of expected runs per nine against an average matchup for each projected starter facing the opposing lineup, with 80 percent bootstrap intervals: Messick minus 0.60, Hagen Smith minus 0.47, Williams minus 0.33, Burke minus 0.16, Cantillo plus 0.18, Kay plus 0.26, Bibee plus 0.41, Martin plus 0.58. — Every starter against his opponent, in runs per nine v an average matchup. Messick holds the biggest edge (−0.60) and Martin the worst (+0.58). Bibee’s +0.41 against this lineup is why Game 3 is Chicago’s best game, and Smith (−0.47) is Chicago’s best matchup on the mound.

IV. Where Chicago can hurt Cleveland

Here is the heart of it. Cleveland is a 56% favourite because of one starter and one reliever. Everything around them has cracks, and the data is specific about where. Five of them follow, each mapped to the Chicago hitter or arm best placed to exploit it.

1. The pitches that leak

Grade every pitch on Cleveland’s staff against the league’s version of the same pitch, starters and the top eight relievers together, and a hit list falls out. The most important entries are the ones thrown often. Joey Cantillo’s four-seamer is a third of his pitches and allows a .369 xwOBA, 40 points worse than the league’s four-seamer, on 224 plate appearances. Gavin Williams’s sinker, 15% of his mix, allows .404. Out of the bullpen, Matt Festa throws his sweeper 40% of the time and Hunter Gaddis his slider half the time; both run 20 to 40 points worse than the league.

Lollipop chart of Cleveland pitches ranked by shrunk xwOBA allowed against the league on that pitch type. Worst: Franco Aleman slider plus 68 points, Messick slider plus 59, Messick curveball plus 51, Williams sinker plus 49 (.404 xwOBA, 15 percent use), Festa sweeper plus 40 (40 percent use), Cantillo four-seam plus 40 (.369, 33 percent use, 224 plate appearances), Cantillo cutter plus 38, Williams cutter plus 34, Bibee curveball plus 30, Bibee sweeper, Herrin slider and Gaddis slider plus 20 to 21. Best, shaded green: Messick changeup minus 67, Messick four-seam minus 66, Sabrowski slider minus 65, Cade Smith splitter minus 56. — Cleveland’s staff, pitch by pitch, against the league. The red lollipops are the targets; size shows how often each pitch is thrown, which is what makes it huntable. The green band is the list to lay off: Messick’s four-seamer and changeup, Cade Smith’s splitter.

Even Messick has a soft underbelly. His slider (.343) and curveball (.319) both grade 50 to 60 points worse than league. They are only 15% of his pitches, on 42 and 36 plate appearances, so this is a lean, not a law. But it says something useful: when he goes away from the fastball-changeup pair, the Sox should be swinging.

2. The attack map

Location tells the same story more sharply. Grid the zone, then compare what Cleveland’s pitchers allow in each cell with what Chicago’s lineup does there, both shrunk toward the league. Where both are positive, Cleveland misses into Chicago’s damage.

Six heat maps in two rows (against right-handed and left-handed pitching) and three columns: what Cleveland's pitchers allow by location, what the White Sox lineup does by location, and the two added together, all relative to the league in xwOBA points, from the catcher's view. Against right-handers the strongest attack cell is just off the third-base edge at belt-to-letters height, plus 96. Against left-handers it is low and just off the third-base edge, plus 107, with a second hot cell high on the first-base side, plus 64. — The attack map, from the catcher’s view. Against Cleveland’s right-handers the hottest cell is just off the plate on the third-base side, belt to letters (+96); with six of Chicago’s nine batting left-handed against righties, that is the pitch drifting away from them. Against the left-handers it is low and off that same edge (+107). Cells rest on a few dozen plate appearances each, shrunk toward the league, so read the clusters, not the single numbers.

3. Load the left side

Cleveland’s right-handers carry real platoon splits, and Chicago’s lineup is built to take advantage. Tanner Bibee allows a .337 xwOBA to left-handed batters and .296 to right-handers, a 40-point gap. Hunter Gaddis, the most-used middle reliever, is worse: .341 against lefties, a 57-point split. Colin Holderman runs 35 points worse against left-handers. Six of the Sox’s projected nine bat left-handed against right-handed pitching.

Dumbbell chart of shrunk xwOBA allowed to left- and right-handed batters for twelve Cleveland pitchers. Biggest gaps: Hunter Gaddis .341 v left-handers against .284 v right-handers; Tanner Bibee .337 against .296; Holderman .323 against .288; Joey Cantillo, a left-hander, .328 v lefties against .299 v righties (a reverse split); Tim Herrin, a left-hander, .325 v right-handers. Messick and Sabrowski show almost no split. — Platoon splits across Cleveland’s staff. Two findings stand out: the right-handed middle relief is vulnerable to left-handed bats, and Cantillo runs a reverse split, worse against left-handers (.328) than right-handers (.299). In Game 4 the platoon advantage belongs to Chicago’s lefties too.

4. The soft middle behind Cade Smith

Cade Smith is the best reliever in this series by a distance: a 2.31 FIP, a .270 xwOBA allowed and a splitter batters can’t touch (.187). The next two, Holderman and Erik Sabrowski, are good too. Behind them it thins out fast. Cleveland’s top three allow a .279 xwOBA between them; relievers four to eight allow .312, almost exactly league average. And it was the middle that pitched in September: Gaddis (14 innings) threw more relief innings than Smith (13). Holderman was activated from the injured list on September 28, having thrown a third of an inning all month.

Two-panel chart of Cleveland's eight relievers ranked by model quality. Left: shrunk xwOBA allowed, top three in navy (Cade Smith .270, FIP 2.31; Holderman .304; Sabrowski .264) and the other five in orange between .304 and .324, around the league line. Right: relief innings since September 1, led by Gaddis 14.0 and Smith 13.0, with Holderman at 0.3. — The soft middle. Get to Cleveland’s starter early and the innings in between belong to relievers who are, on contact quality, league average. That is where Chicago’s left-handed bats should be stacked.

5. Run

This is the cleanest edge in the data. Cleveland threw out 22 of 127 base-stealers this season, 17.3%, against a league rate of 23.0%. That is the fourth-lowest rate in baseball. The split behind the plate is stark. Patrick Bailey threw out 17 of 53 for Cleveland, 32%. Austin Hedges threw out three of 45, under 7%. Bailey started 14 games in September and Hedges 10, so check the lineup card. Then there is the pitcher on the mound. Messick allowed two steals all season in 186⅓ innings, but Cantillo allowed 16 (0.91 per nine) and Williams 14 (0.68 per nine).

Three panels. Caught-stealing percentage: Bailey 32 percent (17 of 53), Hedges 7 percent (3 of 45), Guardians 17 percent (22 of 127), MLB 23 percent. Steals allowed per nine by Cleveland starter: Cantillo 0.91 (16 steals, 7 caught), Williams 0.68 (14, 6), Bibee 0.38 (8, 1), Messick 0.10 (2, 0). White Sox runners' steals and caught: Vargas 22 and 2 (92 percent), Antonacci 20 and 8, Doyle 19 and 6, Luisangel Acuña 18 and 6, Peters 8 and 2. — Run on Cleveland, and pick the night: Hedges catching, Cantillo or Williams pitching. Chicago has the runners to do it. Miguel Vargas stole 22 bases in 24 attempts, Sam Antonacci, Brenton Doyle and Luisangel Acuña 18 to 20 each.

Turning it round: Cleveland’s holes

Attack works in both directions. Take each Sox arm’s best pitch, chosen on whiffs and contact quality, and look up every Guardians hitter against that exact pitch type from that hand. The targets are specific: Bazzana against Grant Taylor’s slider (−55 points) and Smith’s slider (−37); Adell against Taylor’s slider (−44); Angel Martínez against right-handed changeups (−68), which is exactly what José Urquidy and Huascar Brazobán throw. The warnings are as useful as the targets. Bryan Hudson’s four-seamer into Jo Adell (+72) is the matchup to avoid, and Chase DeLauter has hit left-handed sliders hard (+64).

Heatmap of each Guardians hitter's shrunk xwOBA against the league on the specific pitch type of each White Sox arm's best pitch, grouped by pitcher hand and pitch type, with plate appearances. Green holes include Bazzana minus 55 against right-handed sliders (Taylor) and minus 48 against left-handed four-seamers (Hudson), Martínez minus 68 against right-handed changeups (Urquidy, Brazobán) and minus 62 against right-handed sweepers (Hicks, Fedde), and Adell minus 44 against right-handed sliders. Red danger cells include Adell plus 72 against left-handed four-seamers and DeLauter plus 64 against left-handed sliders. — Cleveland’s hitters against the exact pitch types Chicago’s arms lean on. Green is a hole, red a danger, and grey numbers mark cells under ten plate appearances. Two thin rows, Chris Murphy’s curveball and Sean Newcomb’s slurve, are left out because Cleveland’s hitters have barely seen those pitches.

The edges, ranked

Put the whole matchup model to work at the individual level, every Sox hitter against every Cleveland starter and every Sox arm against every Cleveland hitter, and rank Chicago’s advantages.

Two ranked bar charts of Chicago's projected edges in xwOBA points. Bats against Cleveland starters: Vargas v Williams plus 49, Murakami v Cantillo plus 45, Antonacci v Bibee plus 43, Braden Montgomery v Bibee plus 37, Vargas v Cantillo plus 34, Teel v Bibee plus 33. Arms against Cleveland bats: Kay v Bazzana plus 68, Kay v Lowe plus 56, Grant Taylor v Adell plus 56, Hudson v Rocchio plus 50, Smith v Bazzana plus 50. — Chicago’s best individual matchups. At the plate it is Vargas against Williams (+49), Murakami against Cantillo (+45) and Antonacci against Bibee (+43). On the mound, Kay against Bazzana (+68) and Taylor against Adell (+56). The Taylor–Adell matchup is the one Chicago’s manager should save Taylor for.

The plan, game by game

Four game-plan cards. Game 1, Martin v Messick, 41 percent: hunt Messick's curveball (plus 51 points, 8 percent use); stack Vargas plus 19 and Murakami plus 11; Martin attacks Adell plus 10; Messick allowed 2 steals. Game 2, Kay v Williams, 45 percent: hunt Williams's sinker (plus 49, 15 percent); stack Vargas plus 49; Kay attacks Bazzana plus 68, Lowe plus 56, DeLauter plus 37; Williams 14 steals, 6 caught. Game 3, Burke v Bibee, 56 percent: hunt Bibee's curveball (plus 30); stack Antonacci plus 43, Braden Montgomery plus 37, Teel plus 33; Burke attacks Martínez plus 22. Game 4, Smith v Cantillo, 51 percent: hunt Cantillo's four-seamer (plus 40, 33 percent use); stack Murakami plus 45, Vargas plus 34; Smith attacks Bazzana plus 50; Cantillo 16 steals, 0.91 per nine. — The plan, computed. Each card’s pitch to hunt is the starter’s leakiest pitch weighted by how often he throws it. The bats and targets are the top three edges from the matchup model, and the running line is the starter’s own season.

Read across the four cards and the series plan writes itself. Game 1 is the hardest: Messick gives almost nothing and holds runners, so the job is to make him throw the curveball and slider, then get to the middle of the bullpen. Game 2 is the Vargas game. He has a 49-point edge on Williams and the sinker to hunt, and Kay has his two biggest edges of the series in Bazzana and Lowe. Game 3 is the one the model says Chicago should win: load the left side against Bibee’s 40-point split. Game 4 is the running game: Cantillo allows the most steals of the four and throws a four-seamer the league hits, and Murakami has a 45-point edge on him.

V. The bullpens: quality and fatigue

Rank both bullpens best to eighth on the model’s FIP-based quality and they cross. Cleveland is better at the top, with Smith the best arm on either side. Chicago is deeper, rating higher than Cleveland from the fourth arm down. The number that may matter more is fatigue. Chicago’s top eight threw 303 pitches in the seven days before Game 1, more than half of them in Houston. Cleveland’s threw 16.

Left: model reliever quality by rank for both bullpens; Cleveland starts higher (Cade Smith 0.35) but drops below Chicago from the fourth arm down, with the middle innings shaded. Right: pitches thrown by each club's top eight relievers from September 26 to October 2 with days of rest before Game 1: Newcomb 58, Hicks 56 (two days), Taylor 54, Urquidy 50, Smith 38, Brazobán 32 (two days); Cleveland's relievers almost all zero with seven days' rest, Armstrong 11 and Cade Smith 5. — Quality and freshness. Chicago’s best relievers (Taylor, Newcomb, Smith) each threw 38 to 58 pitches in Houston and go into Game 1 on three days’ rest; Hicks and Brazobán on two. Cleveland’s bullpen is almost untouched. The off-day before Game 2 helps Chicago more than Cleveland.

That is the hidden cost of the Wild Card round, and the reason the Fedde-in-Game-4 lever is worth considering: a starter-length arm in Game 4 spares a bullpen that has carried the club for a week.

VI. The trade that didn’t work

In June, this column priced a rumoured deal for Seth Lugo as the answer to a thin rotation. Chicago never made it; Lugo spent the whole season in Kansas City. On August 2 the White Sox went for Luis Castillo instead, sending reliever Seranthony Domínguez, catcher Boston Smith, outfielder Nolan Jones and cash to Seattle. It did not work, and the numbers are blunt about it.

Left: Luis Castillo's 2026 rates in Seattle and Chicago: strikeout rate 19.4 v 17.4 percent, walk rate 7.2 v 8.1, home runs per nine 1.4 v 2.6, ERA 5.1 v 8.4, FIP 4.5 v 6.4. Right: raw ERA shrunk to true-talent ERA with 80 percent intervals: Seattle 5.06 to 4.81, Chicago 8.42 to 6.09, season 5.85 to 5.48, against a starter cohort mean of 4.14. — Castillo before and after the trade, and what the sample implies once it is shrunk toward the starter population. Even the season-long true-talent read (5.48) sits well above the cohort’s 4.14.

In Chicago he made six starts and nine appearances, allowed 29 earned runs in 31 innings and gave up 2.6 home runs per nine. He went on the injured list on September 24 with right-shoulder impingement and won’t pitch in this series. The part the box score hides is that he wasn’t good in Seattle either: a 5.06 ERA and a 4.51 FIP before the trade. Shrink his whole season and his true-talent ERA is 5.48 (80% interval 4.79–6.18), a below-average starter bought at a trough. The saving grace is the price. Of the three players Chicago sent, only Domínguez has 2026 major-league innings in this data: 38 for Chicago with a 3.99 ERA, 12 saves and a 13.9% walk rate, and a 3.04 ERA in 24 innings for Seattle since. Jones and Smith haven’t played in the majors this season. A walk-prone closer and two players outside the big-league picture is a modest bill for a miss, and Chicago’s bullpen still out-rates Cleveland’s from the fourth arm down without him.

What this tells us

The headline number says Cleveland, 56–44, and it has earned that number honestly: the best pitcher in the series, who can pitch twice; the best reliever; home field for three of five; and a fresh bullpen against a tired one. But the shape of the series is more interesting than the number. Messick aside, Cleveland’s staff leaks on pitches it throws a lot: Cantillo’s four-seamer, Williams’s sinker, the middle relievers’ breaking balls. Its right-handers bleed to left-handed bats, and Chicago brings six of them. Behind Cade Smith the bullpen is league average. And with Hedges behind the plate, Cleveland barely throws anyone out.

Chicago’s path is narrow but well marked: steal a game Messick doesn’t pitch in Cleveland, ideally Game 2 with Vargas against Williams, then hold serve at home, where the model already likes Game 3. Load the left side, get into the middle of the bullpen early, and run whenever Hedges is catching. The Astros were the first series win in twenty-one years. This one, on the numbers, is a little harder than a coin flip, and the data says exactly where Chicago should push.

Method. Series model. Series and game probabilities come from the model’s persisted posterior summaries (team_strength, pitcher_strength) and its bracket simulation (playoff_odds, generated 2026-09-30 14:22 UTC, after Wild Card Game 1). The team-only series probability is Chicago’s probability of reaching the Championship Series divided by its probability of reaching the Division Series (47.2%). Per-game probabilities use the model’s Poisson run rates (intercept, offence, defence, home advantage, starter effect), with ties split evenly and the model’s shrink toward 0.5. The park and form terms aren’t persisted, so a single log-rate offset (−0.005) is solved to reproduce 47.2% and then held fixed when the starters are added. Posterior spread uses 20,000 draws of each rating from a Normal matching its published 80% interval, treated as independent. Path, outcome and leverage numbers are exact calculations from the per-game probabilities. The tornado holds one rating at its 10th or 90th percentile while the others vary. Projections. Starters are this column’s projection from rest and usage, not club announcements. Lineups are each club’s nine most frequent starters since September 14. Pitch data. 2026 Statcast via pybaseball, regular season, plus the two Wild Card games for Chicago’s players. xwOBA is expected wOBA on contact and actual wOBA on strikeouts and walks. Hitter and pitcher cells (pitch family × pitcher hand × batter side) are shrunk toward the league’s September value with 60 plate appearances of prior weight, pitch-type cells with 30 and 5×5 location cells with 25. Starter-v-lineup and hitter-v-starter projections combine pitcher and hitter cells by their ratio to league, weighted by usage to each batter side, converted to runs at a wOBA scale of 1.24 and 38.5 plate appearances per nine; intervals are 300 bootstrap resamples of pitches. A Sox arm’s “best pitch” maximises whiff rate minus twice its xwOBA excess over league (minimum 20 plate appearances). A starter’s “pitch to hunt” is his worst pitch against league weighted by usage. Movement is Statcast pfx in inches, arm side positive. Context. Form is the site’s schedule-adjusted run differential. Park factors are home-game over road-game runs per game for each club, 2022–2026; the home-run factor uses 2026 batter game logs only. Reliever quality is the model’s FIP-based rating (reliever_quality). Workload counts each reliever’s pitches from September 26 to October 2, including both Wild Card box scores. Stolen bases, caught stealing and catcher splits are 2026 regular season from the MLB Stats API. Castillo. His true-talent ERA is a Gamma-Poisson posterior on earned runs per inning, with the prior fit to 2026 starters (≥10 GS, ≥60 IP); FIP uses a 2026 constant of 3.19. Results, box scores and transactions are from the MLB Stats API. Data through 2026-09-30.