Michigan-Arizona, Exactly as Predicted. Then Lendeborg Went Down.

By Collin Lee | April 5, 2026

Before the tournament started, the model said Michigan would meet Arizona in the Final Four. Michigan would win, 90-86, at 66% confidence.

Last night, Michigan met Arizona in the Final Four. Michigan won, 91-73.

The model was right about the matchup, right about the winner, and conservative about the margin. What it got wrong was the score - not because Michigan underperformed, but because Michigan was even more dominant than the numbers suggested. Five tournament games, all 90 or more points. The first team in NCAA history to do that, per the tournament broadcast. The model saw a close semifinal. Michigan turned it into a coronation.

And then, in the first half, Yaxel Lendeborg drove to the basket, landed on Arizona center Motiejus Krivas's foot, and crumpled to the floor with a sprained MCL and a rolled ankle. Michigan's First-Team All-American, Big Ten Player of the Year, and the best player on the model's championship pick limped to the locker room. He played 14 minutes total. He hit all three of his three-pointers. And now an MRI says low-grade MCL sprain and a bone bruise, and the championship is tomorrow night.

The model gives Michigan a 98% chance of winning the title. The model doesn't know about the knee.

One of two championship game picks correct. Michigan is in. Duke - our original pick for the other side - was replaced by UConn, who beat Illinois 71-62 in the first semifinal. The model had Illinois at 53% in that game - our closest prediction of the tournament, and it went the other way. UConn's tournament pedigree - third championship game in four years - was the deciding factor in a defensive slugfest where both teams shot under 36%.

The full results

One exact match, one complete miss. Duke and Florida were both eliminated before the Elite 8. Michigan and Arizona made it through five rounds to meet exactly as predicted. The bracket's two halves couldn't be more different - the East/South side has been wrong since the Round of 32, while the West/Midwest side has been a machine.

The game the model predicted

Michigan 91, Arizona 73. An 18-point win. The model had it at 66% with a 4-point margin. The model was right about the winner and wrong about how easy it would be.

Michigan came out like they'd already seen the game film. They forced four Arizona turnovers in the first five minutes, scoring seven points off them. Trey McKenney threw down a fast-break dunk to make it 16-5. By the time Arizona called their first timeout, Michigan led 26-10 - Arizona's largest deficit of the entire season, and it was only 10 minutes into the game.

Then Lendeborg went down.

Two quick fouls, a drive to the basket, a bad landing on Krivas's foot. MCL sprain. Ankle roll. He walked to the free throw line, hit both free throws, and walked to the locker room. Michigan's best player, the one the model's entire championship prediction is built around, was gone.

It didn't matter. Aday Mara, the 7-foot-2 junior from Spain, took over with 26 points on 11-of-16 shooting. Elliot Cadeau ran the offense with 13 points and 10 assists. Michigan led 48-32 at halftime without their star. They shot 9-of-13 to start the second half. The lead hit 30. Arizona's Tommy Lloyd, whose team had won 36 games this season, said afterward: "No one has been able to do that to us all year."

Lendeborg came back for the second half wearing a compression sleeve. He hit three three-pointers in a row - 3-for-3 from deep on a bad knee - then sat on the stationary bike for the final five minutes while his teammates finished the job. Fourteen minutes. Eleven points. All on one good leg.

The model said Michigan 90, Arizona 86. Michigan won 91-73. The model was right about Michigan. It wasn't right enough.

The game nobody predicted

UConn 71, Illinois 62. Our original bracket had Duke vs. Florida in this semifinal slot. Neither team made it past the Elite 8 and Sweet 16, respectively. In their place: a UConn program going for its third title in four years and an Illinois team making their first Final Four since 2005.

In the Elite 8 post, I applied the model to this new matchup and it gave Illinois a 53% edge - the closest prediction of the entire bracket. UConn won by 9. The model's margin of error on a coin-flip prediction doesn't bother me. What bothers me is the game itself: both teams shot under 36% from the field. Illinois went 6-of-26 from three. It was an ugly, grinding, defensive masterpiece, and UConn's experience in these moments - 19 consecutive NCAA Tournament wins from the Sweet 16 or later - was the difference.

Braylon Mullins, the freshman who killed Duke with a 35-foot three in the Elite 8, scored 15 points including a dagger three with 52 seconds left after going scoreless in the second half. Tarris Reed Jr. had 17 and 11. UConn is in the championship game. Again.

The variable the model can't see

Here's where I stop being analytical and start being honest.

The model gives Michigan a 98% chance against UConn. Ninety-eight percent. That's the highest championship confidence it has produced in this entire series - higher than the 80% against Duke, higher than the 94% against Illinois. By the numbers, Michigan should walk through this game.

But the model doesn't know that Lendeborg can barely walk.

This is the limitation we've been talking about since the very first post. The model can't account for injuries announced the morning of a game. It can't read a team's body language in warmups. It processes 39 features per matchup and none of them are "star player has a bone bruise in his knee." The 98% is a prediction for a healthy Michigan team. Tomorrow night's Michigan team may not be healthy.

Lendeborg says he's playing. "I'm gonna play unless I can't walk at all," he told reporters. The post-game MRI came back clean - low-grade MCL sprain, small bone bruise. He went through Sunday practice. But he played 14 minutes last night when he normally plays 30+. He was on a stationary bike for the last five minutes of a Final Four game. "Clean MRI" and "ready to play 35 minutes in a championship game" are not the same thing.

Michigan won by 18 without him last night. Against Arizona. UConn is not Arizona. UConn is a program that's won two of the last three national championships. UConn is the team that came back from 19 down against Duke and won on a logo three. UConn is coached by Dan Hurley, who called this "a life-and-death struggle." Michigan's depth carried them through one game without their star. Whether it carries them through a championship against UConn's championship DNA is a different question.

The model sees 98%. I see a player on a stationary bike.

What the model says about the championship

From the original bracket (predicted pre-tournament)

The original bracket predicted Michigan over Duke, 80%, 79-78. Duke is gone. This prediction died in the Elite 8. But the model's core belief - Michigan wins the championship - is still standing.

New prediction (same model, actual matchup)

Michigan over UConn, 98%, 83-69. The model's most confident championship prediction yet. But it was calculated on season-long data with a healthy roster. The Lendeborg injury happened last night. I have no way to adjust for it - the model doesn't take injury inputs - and I'm not going to fabricate a number.

So here's what I'll say instead: the 98% is the model's honest assessment of Michigan vs. UConn at full strength. How much Lendeborg's knee changes that equation is the one question I can't answer with data. If he plays at 80%, maybe it's still 90%. If he's limited to spot minutes, maybe it's a coin flip. I genuinely don't know, and any number I gave you would be a guess dressed up as analysis.

The model has been right about Michigan all tournament. Five games, five wins, all by double digits. Michigan has now scored 90+ in every tournament game - a feat no team in NCAA history had accomplished before this run. The roster depth that carried them past Arizona without Lendeborg is real. Aday Mara can dominate. Cadeau can run the show. McKenney can score in bunches.

But 98% assumes the best player is the best player. Tomorrow night, he might be a guy with a sleeve on his knee.

What's next

Monday night. Michigan vs. UConn. 8:30 PM ET. Lucas Oil Stadium in Indianapolis.

The model's champion pick versus the program that's won two of the last three championships. A healthy Michigan wins this going away - the model is emphatic about that and five games of evidence support it. A diminished Michigan against this UConn team is something else entirely.

This whole series started with a question: what happens when you remove emotion from your bracket? Six posts later, I have part of an answer. The emotion-free bracket got the champion right. It got the semifinal matchup right. It got 43 of 62 games right. It was wrong about Duke, wrong about Florida, wrong about half a dozen games where the math said one thing and the tournament said another.

But Michigan - Michigan it got right from the start. Five rounds. Five wins. One more to go.

The model says 98%. The knee says we'll see.

Tomorrow night.

Update: Michigan won. The final chapter.