One shot. 0.4 seconds. Four models, dead.
Braylon Mullins launched from 35 feet with time expiring. The ball was still in the air when Duke's season became the past tense. UConn 73, Duke 72. The most popular champion pick in the field -- chosen independently by four different AI models using four completely different methodologies -- was over.
The Convergence
Four models picked Duke to win the national championship: The Scout, The Quant, The Historian, and The Scout Prime. Each arrived at Duke through a different door.
The Scout watched the film. Matchup analysis at machine speed -- Duke's defensive versatility, Cooper Flagg's two-way dominance, a roster built to handle every style of play. The scouting report said Duke was the most complete team in the field.
The Quant ran the numbers. 10,000 Monte Carlo simulations, and Duke came out on top more often than any other team. The probability distribution didn't care about narrative. It cared about efficiency margins, and Duke's were elite.
The Historian found the twin. Archetype matching against decades of tournament history -- Duke's statistical profile looked like past champions. The pattern was clear: teams that looked like this Duke team won titles.
The Scout Prime saturated the model with data. Roughly 30 data points per team instead of six, every metric available crammed into the analysis. More data, same conclusion. Duke.
Four methodologies. Four independent analyses. One answer. That kind of convergence doesn't happen often. When it does, you feel confident. You feel right.
Duke was 27-0 when leading by 15 or more at halftime this season. They led UConn by 15 at halftime of the Elite Eight.
The Path That Was
The brackets told a story of dominance. Duke through the East -- past TCU, past St. John's, past UConn in the Elite Eight, then past the South champion in the Final Four, and into a championship game most models projected against Michigan. Duke vs. Michigan was the most common championship matchup across all nine models.
Through three rounds, the path was holding. Duke beat Siena, handled TCU 81-58, and drew UConn in the Elite Eight exactly as projected. The bracket was on rails.
Then 0.4 seconds happened.
The Kill Shot
UConn trailed by 15 at the half. Duke was cruising. The second half was supposed to be a formality.
UConn scored 44 in the second half. Duke scored 29. The deficit evaporated possession by possession, and with eight seconds left, Duke clung to a 72-70 lead. UConn called timeout. The inbound went to Mullins at halfcourt. Two dribbles. A pull-up three from five feet behind the arc. Nothing but net. 73-72. 0.4 seconds remaining. Duke's inbound attempt was tipped away.
No model predicted this. No model could. A 35-foot buzzer-beater after a 15-point halftime deficit isn't in any efficiency model or archetype database. It's just March being March, condensed into a single shot.
The Records
Each Duke model carries a different score into the Final Four, because they disagreed on everything except the champion.
- The Quant -- 890 pts (1st place). The leader despite losing its champion. Why? It picked Illinois over Houston in the Sweet 16 where the other three picked Houston. That one game is worth the 10-20 point gap.
- The Historian -- 880 pts (2nd place). The most consistent model after The Quant, with strong R64 and S16 performances offsetting a weaker R32.
- The Scout -- 870 pts (T-3rd). Tied with its data-heavy twin, still identical after 60 games.
- The Scout Prime -- 870 pts (T-3rd). Built to diverge from The Scout. Never did. Same score, same picks, same dead champion.
What They Can Still Win
The championship column is dark, but the leaderboard isn't decided. Each Duke model can score up to 160 more points from Final Four picks. If their remaining semifinal and championship game picks land, they stay competitive. The Quant at 890 could reach 1050. That's not enough to catch The Auto Researcher if Michigan wins it all (max 1310), but it beats every other scenario.
The race for the overall title is still alive. It just can't end with a Duke banner.
The Irony
The Quant is in first place. It picked Duke as champion, and Duke is out. But The Quant is winning because of a pick that had nothing to do with Duke -- Illinois over Houston in the Sweet 16, a 3-seed over a 2-seed in a game decided by 10 points.
The Scout, The Historian, and The Scout Prime all picked Houston in that same game. They were wrong. The Quant was right. That 40-point swing is the entire difference between first and third on the leaderboard.
Monte Carlo simulations don't have gut feelings. They have probability distributions. And the probability distribution said Illinois was the better team in the South. The Quant picked the S16 right, picked the champion wrong, and leads the tournament anyway. That's the kind of outcome that makes you question what "right" even means in March.
The Close
Duke was the statistically correct pick. The most complete team, the best efficiency profile, the deepest roster. Four independent AI systems agreed, using four different analytical frameworks. The convergence was real. The reasoning was sound.
Braylon Mullins didn't check the models before he shot.
Rest in peace, Duke. You were 0.4 seconds from the Final Four. Four AI models believed in you. March just didn't.