Arms vs. hot streak, revisited: the hot streak took it

Oklahoma is the national champion. The Sooners took the College World Series final two games to one, closing it out in Omaha with a 13-2 rout in Game 3, and they did it in almost exactly the way our model said they could, if they could. It is Oklahoma’s first national title since 1994, ending a 32-year drought.

We went into this series with North Carolina as the favorite. Not a heavy one, but a clear one: 58.1 percent to win the best-of-three, a number that had held since the bracket was drawn and then firmed up once the simulation started accounting for who was actually on the mound. The case was pitching, and it was a good case. It was also, in the end, the wrong side of a coin the model always said could land either way. (Here is exactly what we said going in.)

This is the recap we owe you, because showing the kitchen means showing the meals that don’t come out the way the recipe predicted.

What it looked like going in

The pre-series argument was simple and, on paper, hard to dispute. North Carolina ran the better arm to the mound in every projected matchup. The Tar Heels’ worst projected starter, McDuffie at a 3.26 ERA, still graded ahead of Oklahoma’s best, Wesloski at 3.63. There was no game in the series where the model saw Oklahoma taking the mound with the pitching edge, and in a short series, pitching depth is the thing that usually decides it. That is why the per-game numbers all leaned North Carolina, 54 to 56 percent, and why the series sat at 58.1.

UNC favored in all three games — but never by more than 56%. A thin, coin-flippable edge.

One word in that paragraph is doing more work than it looks like: projected. The model lined up each team’s rotation at its season ERA and assumed those arms would be available, rested, in the roles the season suggested. Hold onto that, because it is where the series turned.

Against that, the model gave Oklahoma a real but minority path, and it named exactly what that path looked like. The Sooners were the hottest team in Omaha, 9-1 to get here after entering the bracket as the longest shot in their group. They missed more bats than North Carolina, 10.4 strikeouts per nine to 9.4. And the live scenario, in the model’s own words going in, was that Oklahoma rides the hot streak one more weekend, the offense stays as hot as it has been, and the bats beat the arms before the pitching depth can settle the series.

That is not hindsight. That was the 42 percent the model gave Oklahoma going in. And it came in.

How the series actually went

North Carolina did its part for two games’ worth of the projection. The series reached 1-1, with the Tar Heels forcing a winner-take-all Game 3, which is precisely the shape the model assigned the highest probability: a competitive best-of-three that comes down to one final night. Through two games, the careful 58-42 lean looked like it was tracking.

Then Game 3 went sideways for the favorite, and it went sideways in the one place the model could not see coming, because it wasn’t on the projection sheet.

The innings the model couldn’t count

Here is the thing a season ERA cannot tell you: how many innings the arm behind it has already thrown to get to tonight.

North Carolina did not arrive at this Game 3 fresh, and the way it got here is the whole story. The road to a national title runs through a conference tournament, a regional, a super regional, and then a multi-week College World Series gauntlet, a long sequence of elimination and near-elimination games that all draw from the same finite staff. By the time the Tar Heels reached a winner-take-all final, the rotation that looked so deep on paper in early June had been used, and used hard, just to survive the rounds before it.

Trace the arms across the series and the depletion is right there in the pitch counts. North Carolina opened the final with Jason DeCaro and went to Walker McDuffie in relief, spending two of its better arms in a Game 1 loss. With its season on the line in Game 2, starter Ryan Lynch left with an apparent injury early, and the Tar Heels turned to Caden Glauber, their best and most trusted arm, who threw 65 pitches across five shutout innings to force a Game 3. That is exactly the right call to extend a season. It is also exactly the kind of decision that empties a staff. By the time Monday’s winner-take-all arrived, North Carolina had already spent its front-line starters and burned its ace in relief the day before.

So the Tar Heels did the only thing a depleted staff can do: they handed the biggest game of the year to the freshest available arm, a freshman who had been pitching well out of the bullpen, and hoped to cobble together nine innings behind him. When that did not hold and they went back to Glauber, there was nothing left. He faced two batters, both reached, and his night was over after eight pitches. The front-line rotation depth that was the entire basis for North Carolina’s edge was, by the final night, an injured starter, a spent ace, and a bullpen being asked to do what a rested rotation was supposed to do.

That is the gap our pre-series model could not price, and it is a bigger gap than any single decision. The simulation matched up starters by their season ERA, the rested-arm, full-season number. A 2.13 does not get tired in a spreadsheet. But it gets tired in Omaha, in 100-degree heat, across a tournament that asks a pitching staff to keep answering the bell long after the rotation’s front end has emptied out. North Carolina’s edge was depth, and depth is exactly the asset the grind erodes first, because depth is what you burn to survive the rounds before the final. The model saw the season ERAs. It could not see the innings ledger behind them.

The bats the model warned about

Oklahoma’s offense by game. Glauber’s Game 2 relief held the bats to 2 runs — the only time all series. Once North Carolina’s staff was spent, the Sooners erupted for 13.

While North Carolina’s pitching plan came apart, Oklahoma did the thing the model flagged as their entire path to the title. The Sooners hit, and they hit relentlessly: 14 hits, 13 runs, and a 7-for-16 line with runners in scoring position, which is the number that turns a good offensive night into a blowout. Kyle Branch drove in six, including a three-run home run in the eighth that put the game away. Dalton Tockey homered. Jaxon Willits reached base four times. This was not a team stealing a low-scoring game behind a strikeout arm, which was one version of the upset the model imagined. It was the other version, and the more emphatic one: the juggernaut offense simply overwhelming a pitching staff that was, all year, among the best in the nation.

That is the part worth sitting with. North Carolina’s arms were not a mirage. They were genuinely excellent, which is why the model trusted them. Oklahoma beat them anyway, because a hot offense in a single game is the one force that can make the best pitching staff in the country look ordinary. The model gave that outcome 42 percent. Four times in ten, the bats win. This was one of those times.

A 2.13 does not get tired in a spreadsheet. But it gets tired in Omaha.

What the model got right, and what it didn’t

It would be easy to write this as a miss, and in the narrowest sense it was: we favored the team that lost. But the more honest read is that the model called the structure of this series almost exactly. It identified the fault line correctly, arms against offense. It assigned real probability to the offense winning and described, in advance, the precise scenario in which it would. And it nailed the shape, a competitive series going the full three games to a winner-take-all Game 3.

What it could not do was count the innings. The model priced North Carolina’s rotation at its rested, full-season ERAs, and by the final night of a conference-tournament-plus-CWS gauntlet, that rotation was no longer rested. That is not a flaw in the run-production model. It is the boundary of what a pre-series simulation can know, and it is a good argument for exactly what we said going in: you keep the simulation running after every pitch, because the gap between the rested arm on paper and the spent staff on the mound is where short series, at the end of a long tournament, are won and lost.

Oklahoma was the team nobody’s numbers believed in, ours included. They went 9-1 to reach the final, and then won the one game that mattered most by doing the one thing we said could beat the favorite. Arms against the hot streak. This time, the hot streak took it. Congratulations to the Sooners.