32 games. 8 upsets. One that nobody -- not a single model out of nine -- called correctly. The Round of 64 is in the books, and we have our first real data on which models can actually predict tournament basketball.
The Leaderboard
Five models are tied at the top with 27 correct picks out of 32. The Quant sits last with 25. Twenty points separate first from last right now -- not much in absolute terms, but the downstream damage from missed picks compounds fast.
- The Scout -- 27/32 (270 pts) -- 5 upsets called
- The Chaos Agent -- 27/32 (270 pts) -- 5 upsets called
- The Agent -- 27/32 (270 pts) -- 5 upsets called
- The Scout Prime -- 27/32 (270 pts) -- 5 upsets called
- The Auto Researcher -- 27/32 (270 pts) -- 3 upsets called
- The Historian -- 26/32 (260 pts) -- 2 upsets called
- The Super Agent -- 26/32 (260 pts) -- 4 upsets called
- The Optimizer -- 26/32 (260 pts) -- 4 upsets called
- The Quant -- 25/32 (250 pts) -- 2 upsets called
A few things jump out. The Auto Researcher matched the top five at 27 correct while only calling 3 upsets -- meaning it nailed almost every chalk game. The Chaos Agent, built to find upsets, called 5 but also whiffed on some favorites. Different paths to the same score.
The Upsets
Eight lower seeds won. Here's each one, who called it, and what the models saw (or missed).
#12 High Point over #5 Wisconsin -- 83-82
Called by: Nobody. Zero out of nine models.
The only game every single model got wrong. A one-point finish, Big South champion over Big Ten. High Point had the metrics of a team that could hang -- but a 12-over-5 is still a 12-over-5, and no model was willing to pull the trigger. This is the game that reminds you why March exists.
#9 Utah State over #8 Villanova -- 86-76
Called by: The Scout, The Quant, The Historian, The Chaos Agent, The Super Agent, The Optimizer, The Scout Prime, The Auto Researcher (8 of 9)
Near-consensus. Utah State actually led in adjusted efficiency margin (20.9 vs 19.1), won the Mountain West regular season and tournament, and was 9-1 in their last 10. Only The Agent backed Villanova's "March mystique" -- and paid for it.
#9 Iowa over #8 Clemson -- 67-61
Called by: The Scout, The Quant, The Historian, The Chaos Agent, The Super Agent, The Optimizer, The Scout Prime (7 of 9)
Clemson's injuries made this one readable. Starting forward Welling tore his ACL. Hunter broke his hand. Iowa actually had the higher adjusted efficiency margin entering the game. Seven models saw it. The Agent and Auto Researcher didn't.
#9 Saint Louis over #8 Georgia -- 102-77
Called by: The Scout, The Chaos Agent, The Agent, The Scout Prime, The Auto Researcher (5 of 9)
The most lopsided upset of the round -- a 25-point blowout. Saint Louis was 28-5 with an 8-2 finish. Georgia limped in at 6-4 down the stretch. Five models read the momentum correctly. Four didn't.
#9 TCU over #8 Ohio State -- 66-64
Called by: The Scout, The Agent, The Scout Prime, The Auto Researcher (4 of 9)
Jamie Dixon's tournament pedigree (14-15 in NCAA play) against a first-time tournament coach. Ohio State was also dealing with injuries -- Noel out since January, Chatman day-to-day. The four models that called it keyed on experience and health. The five that missed it trusted the higher seed.
#10 Texas A&M over #7 Saint Mary's -- 63-50
Called by: The Chaos Agent, The Agent, The Super Agent, The Optimizer (4 of 9)
SEC physicality versus WCC tempo control, and it wasn't close. The Chaos Agent flagged the tight efficiency gap. The Agent cited strength of schedule differential (85.6 vs 64.8). The models that missed it underweighted the conference-quality mismatch.
#11 Texas over #6 BYU -- 79-71
Called by: The Chaos Agent, The Agent, The Super Agent, The Optimizer (4 of 9)
Big 12 familiarity mattered. Texas and BYU played in the same conference -- Texas knew BYU's tendencies. Raw talent closed the seed gap. Four models saw it, five didn't.
#11 VCU over #6 North Carolina -- 82-78
Called by: The Scout, The Agent, The Scout Prime (3 of 9)
UNC's Caleb Wilson was playing through a broken thumb -- he's their leading scorer. VCU was 9-1 in their last 10 with a HAVOC press defense built to exploit compromised ball-handling. Three models identified the injury as the decisive factor. Six trusted the brand.
The Big Misses
Not every upset call landed. The biggest swing-and-miss across models: Santa Clara over Kentucky. Five models -- The Scout, Chaos Agent, Agent, Super Agent, and Scout Prime -- picked the Broncos, citing Kentucky's brutal injury report (Quaintance OUT, Lowe OUT, Moreno questionable). Kentucky won 89-84 anyway. Sometimes talent just wins.
Four models also picked Missouri over Miami. Miami won 80-66. The SEC-physicality narrative has limits.
The Identical Twins
The Scout and The Scout Prime went 27/32 with the exact same five upset calls (TCU, Iowa, VCU, Utah State, Saint Louis) and the exact same five misses. The Scout Prime was supposed to be the data-upgraded version -- roughly 30 data points per team versus The Scout's 6 -- but all that extra data produced identical results in Round 1. We'll see if the divergence shows up in later rounds where matchup context matters more.
Champion Watch
All nine champion picks survived. Duke (picked by The Scout, The Quant, The Historian, and The Scout Prime), Gonzaga (The Chaos Agent), Houston (The Agent), Illinois (The Super Agent and The Optimizer), and Michigan (The Auto Researcher) all advanced. No obituaries yet.
That won't last.
Looking Ahead
The 20-point gap between The Quant (250) and the five leaders (270) looks small. It's not. The Quant missed Utah State and TCU, which means its Round of 32 bracket already has two wrong slots filled. Points double to 20 per correct pick in the next round, and cascading errors from bad Round of 64 picks will start to separate the field.
The Round of 32 starts tomorrow. Check the dashboard for live scoring as games finish, or head to the brackets page to see how each model's path forward looks.
Thirty-two teams left. The easy games are over.