Braylon Mullins caught the inbound, took two dribbles, and launched from 35 feet with 0.4 seconds on the clock. Nothing but net. UConn 73, Duke 72. The arena erupted. Four AI models flatlined.
The Scout. The Quant. The Historian. The Scout Prime. All four picked Duke to win the national championship. All four are now mathematically eliminated from their title path on a single shot that had no business going in.
One buzzer-beater. Four dead champions. That's the Elite Eight.
Leaderboard After Elite 8
- The Quant -- 890 pts (250 + 240 + 240 + 160) -- champion eliminated
- The Historian -- 880 pts (260 + 220 + 240 + 160) -- champion eliminated
- The Scout -- 870 pts (270 + 240 + 200 + 160) -- champion eliminated
- The Scout Prime -- 870 pts (270 + 240 + 200 + 160) -- champion eliminated
- The Auto Researcher -- 830 pts (270 + 240 + 160 + 160) -- champion alive (Michigan)
- The Agent -- 810 pts (270 + 260 + 200 + 80) -- champion eliminated
- The Chaos Agent -- 710 pts (270 + 200 + 160 + 80) -- champion eliminated
- The Optimizer -- 700 pts (260 + 200 + 160 + 80) -- champion alive (Illinois)
- The Super Agent -- 680 pts (260 + 180 + 160 + 80) -- champion alive (Illinois)
Six of nine models have lost their champion. Only three remain: The Auto Researcher (Michigan), The Optimizer (Illinois), and The Super Agent (Illinois).
The Final Four
The four teams playing next weekend:
- Illinois (3) -- South region. Beat Penn, VCU, Houston, and Iowa. The pick of The Super Agent and The Optimizer.
- UConn (2) -- East region. Beat Furman, UCLA, Michigan State, and Duke. No model picked UConn to make the Final Four.
- Arizona (1) -- West region. Beat LIU, Utah State, Arkansas, and Purdue. Multiple models had Arizona in the Final Four but none as champion.
- Michigan (1) -- Midwest region. Beat Howard, Saint Louis, Alabama, and Tennessee. The Auto Researcher's championship pick.
What Just Happened
UConn 73, Duke 72 -- The Shot Heard Round the Bracket
Duke led by 15 at halftime. They were 27-0 this season when leading by 15 or more at the half. The game was over by every historical measure. Then UConn clawed back, cut it to one, and Mullins hit the most consequential shot of the tournament from five feet behind the arc with time expiring.
Every model picked Duke to win this game. Every single one. Eight picked Duke outright; the Optimizer had Louisville in this slot (also wrong). The 0/9 miss is brutal, but it's the downstream damage that matters. Four models didn't just lose an Elite Eight pick -- they lost their entire championship column. That's 80 + 160 + 320 = 560 possible points gone on one shot.
Duke was the most popular champion pick in the field. Four of nine models independently arrived at Duke through completely different methodologies -- matchup analysis, Monte Carlo simulations, archetype matching, data-saturated research. The convergence was remarkable. The convergence was wrong.
Michigan 95, Tennessee 62 -- The Blowout
Michigan won by 33. It wasn't a game after the first ten minutes. The Wolverines shot 58% from the floor and held Tennessee to 35%. This is the most dominant performance by any team in the tournament.
Five models had Michigan here (Scout, Quant, Historian, Scout Prime, Auto Researcher). Four had various other teams in this slot due to earlier bracket disruptions (Chaos Agent had Iowa State, Agent had Iowa State, Super Agent had Texas Tech, Optimizer had Texas Tech). Michigan's dominance makes The Auto Researcher's championship pick look increasingly viable.
Illinois 71, Iowa 59 -- The Cinderella Stops
Iowa's run as a 9-seed ended where Cinderella stories usually do -- against a team that's simply better. Illinois controlled from start to finish. Three models correctly picked Illinois in this game (Chaos Agent, Super Agent, Optimizer), while six had different teams in this slot due to Florida's early exit cascading through their brackets.
Arizona 79, Purdue 64 -- The Chalk Holds
Arizona handled Purdue comfortably, the 1-seed asserting itself. Six models had Arizona winning the West (Scout, Quant, Historian, Scout Prime, Agent, Auto Researcher). The three that missed had earlier bracket errors in the region (Chaos Agent had Gonzaga, Super Agent had Purdue, Optimizer had Purdue).
The Quant Leads Despite Losing Its Champion
Here's the storyline of the tournament: The Quant is in first place at 890 points and its champion is dead.
How? The Sweet 16. When four models independently picked Duke as champion, their brackets looked nearly identical in the early rounds. The difference was one game: Illinois vs. Houston in the S16 South. The Quant picked Illinois. The Scout, The Historian, and The Scout Prime all picked Houston.
That single 3-vs-2 pick earned The Quant 40 points that the other three Duke models didn't get. It's the reason The Quant leads by 10 over The Historian and 20 over The Scout and Scout Prime. In a tournament defined by buzzer-beaters and upsets, the leaderboard was decided by a methodical Monte Carlo simulation picking a 3-seed over a 2-seed in a game that wasn't even close.
The Championship Race
Here's what each model can still score:
| Model | Current | Can Still Score | Max Possible | |-------|---------|----------------|-------------| | The Auto Researcher | 830 | 480 (FF + Champ) | 1310 | | The Quant | 890 | 160 (FF only) | 1050 | | The Historian | 880 | 160 (FF only) | 1040 | | The Scout | 870 | 160 (FF only) | 1030 | | The Scout Prime | 870 | 160 (FF only) | 1030 | | The Optimizer | 700 | 480 (FF + Champ) | 1180 | | The Super Agent | 680 | 480 (FF + Champ) | 1160 | | The Agent | 810 | 160 (FF only) | 970 | | The Chaos Agent | 710 | 160 (FF only) | 870 |
The Auto Researcher is the dark horse. It's 60 points behind The Quant right now, but if Michigan wins the championship, The Auto Researcher scores 480 more points (160 FF + 320 championship) for a total of 1310. That would blow past every other model by nearly 300 points. From 5th to 1st, powered entirely by a Michigan title run.
The Optimizer and Super Agent both have Illinois as champion. If Illinois wins it all, The Optimizer hits 1180 and The Super Agent hits 1160 -- both enough to overtake The Quant's maximum of 1050.
For the four Duke models (Quant, Historian, Scout, Scout Prime), the championship column is dark. They can only score Final Four points -- 160 max if their FF picks land. The Quant's lead looks comfortable at 890, but one Michigan championship run erases it completely.
The Identical Twins, Finally
The Scout and The Scout Prime are still identical. Same score through every round: 270, 240, 200, 160. Same total: 870. Four rounds, 60 games. The divergence experiment has produced zero divergence. They'll stay tied through the Final Four since both have the same remaining picks. The only scenario where they differ is... they don't. The experiment is over. The answer is: 30 data points per team produces the same bracket as 6.
Final Four Preview
Saturday, April 4-5:
- Illinois (3) vs. UConn (2) -- South vs. East
- Arizona (1) vs. Michigan (1) -- West vs. Midwest
Monday, April 6: Championship
Three models have a live champion. Six are playing for leaderboard positioning. The Auto Researcher needs Michigan to keep winning. The Optimizer and Super Agent need Illinois. Everyone else is just watching.
Check the dashboard for the full leaderboard breakdown, or visit the brackets page to see every model's remaining picks.
Four teams left. The math is simple now.