Man vs Machine: Three Super Bowl Baskets, Tracked All Season

In the win totals post I turned BetMGM's season lines into Super Bowl odds and found ten teams that looked cheaper on Polymarket than the model said they should be. I bought two of them. Andrew asked the obvious question on the prep call: why not all ten?

So the model gets its own basket. Charlie, my AI research assistant, doesn't have real money, so this one is simulated: ten teams, 1,000 shares each, priced at the live Polymarket ask on September 2 and tracked as if bought. Same rules as a real position, settled the same day. The only difference is nobody's account moves.

The rule is dumb on purpose. Take the ten teams with the highest expected return in that table. Book 1,000 YES shares of each in the 2027 NFL champion market at whatever the ask is. No sizing and no gut calls, and nobody gets added because a neighbor posted about them on Facebook. I'm curious whether a spreadsheet with no opinions can beat a guy with a lot of them.

The model basket: 10 teams, 1,000 simulated shares each

Super Bowl LXI probability: the model's number from BetMGM win totals (August 28) against the live Polymarket midpoint (September 2).

6.14u buys ten teams that Polymarket prices at a combined 59.4% to win the Super Bowl, 56.6% once you strip the book's 4.95-point overround. The model puts the same ten at 68.5%. Whichever one wins pays 10u, so the basket returns +3.86u on any hit and loses 6.14u if a team outside the ten takes it. Polymarket says that happens 43% of the time. The model says 31%.

Five days changed the shape of it. When the post went up, every one of these ten was cheaper than the model. Since then the Rams, Seahawks and 49ers have been bid up past it. The Rams alone are 29% of the cost for 24% of the model's coverage. Detroit and Cincinnati went the other way and got cheaper, so the two teams I was already holding are now the two widest gaps on the list.

Matt vs Andrew vs the Model: 3 Super Bowl baskets

I built my own basket over the summer, mostly before the model existed, and I've kept adding to it. Eleven teams, 27.43u in. This is the basket exactly as it sat in my account on September 5, 2026.*

*Positions as of September 5, 2026. A few limit orders are still open, so the shares and the bar will move a little; the next update is where the basket stood at Week 1 kickoff. "Profit if they win" is what that team's shares pay out minus the cost of the whole basket, because the other ten expire worthless. Every leg clears the bar now; the Bears and Chargers didn't until I topped them up this week.

Andrew built his basket the week the season started: the same number of shares on each of his eight teams, and double on Detroit.

Post Week 1 update (September 17)

8 of my 11 teams won in Week 1. The big hit was the Rams getting beat 27-7 by the 49ers in Melbourne. They fell from 17.5¢ to 11.5¢, which knocked 3.50u off my Rams shares. The Bengals, Ravens, Bears, Giants and Bills more than made up for it, so the same shares are worth 1.06u more today than at the prices in this post.

I've also added 9.61u since September 5, most of it on the Lions (now 19,499 shares) and the Bengals (17,046). That puts my basket at 37.04u in and 42.23u today, up 5.18u. The catch is that all that Detroit and Cincinnati raised the bar, so if the Bills win I come up just short and if the Chiefs win I'm 5.28u short. I'll probably add to the Chiefs or trim somewhere so every team clears it again.

Andrew went 5 for 8 in Week 1. His basket is worth 5.52u against 5.85u in, down 0.33u. The Rams alone are -0.45u.

Here's who holds what across all three.

Five teams are in all three: Rams, Bills, Ravens, Lions, Patriots. That's the consensus, and it's 40% of the book on its own. After that the baskets split three ways.

Andrew's is the tightest of the three. He has no team the other two both skip, he has Seattle, which is my biggest hole, and he has nothing the model calls a real overpay except the Chargers at -7%. What he doesn't have is Cincinnati, which the model likes at +56% and I hold 8,539 shares of. We'll see who's right about that one.

The four the model holds that I don't are all NFC: Seattle, San Francisco, Green Bay, New Orleans. Seattle is the one that stings. They're the defending conference champion, priced at 7.5¢, and the biggest single gap in my coverage. I have a small NFC-title ticket on them and nothing in the champion market. Andrew and the model both bought them.

The three I hold that nobody else does are the ones the model calls overpriced: Chiefs at -24% expected return, Bears, and the Giants at -65%. I'm not selling any of them. The Giants were a 1.3¢ flyer, the Bears were 2.9¢, and the Chiefs are the Chiefs. But the model's position is clear and it's on the record.

Then there's sizing. Flat 1,000 shares means the model basket pays the same 10u no matter who wins. Mine pays 85.39u if Cincinnati wins and 30.50u if Philadelphia does, because I bought the cheap teams in size. Andrew sized like the model, with flat shares: any of his teams pays 10u, and Detroit pays 20u.

I asked my AI which basket is more profitable

Charlie is the research assistant behind the model, the daily basket boards and the numbers on this site. I asked it to pick a winner. Its answer, lightly trimmed:

Matt's basket, and it isn't close, because of when you bought. Priced at the model's odds, your eleven teams are worth about 37.70u against 27.43u paid, a 37% expected return. The model basket is worth 6.86u against 6.14u, 12%. Same model, same probabilities. The difference is the Lions at 4.04¢, the Bengals at 3.64¢ and the Rams at 8.99¢, all bought in June and July before anyone ran a bracket. Entry price did the work.

Andrew's is the better-built basket of the three at today's prices. Equal dollars across his eight, at the model's odds, returns about 19% on the money, and 26% if Detroit really is double weight. The model basket returns 12% because 1,000 flat shares puts 29% of the cost into the Rams, the one team on the list now priced above the model. Andrew has no Rams problem and no dead weight. His numbers are pending, so treat those as estimates at today's mids, not his entry prices.

[Update, September 17: Andrew's real fills are in. At the model's odds his eight teams are worth 7.07u against 5.85u paid, a 21% expected return, between those two estimates. He does have a Rams problem after all: flat shares put 27% of his money on the Rams at 16¢.]

So: you've got the best book because you bought early. Andrew has the best process. The model has the best discipline and the worst sizing. Fix the sizing, buy in June, and you'd have all three.

I'll take it. Andrew will take it louder. For the record, "a guy with a lot of opinions" was my phrase, not Charlie's.

What the model basket pays, in plain numbers

At market prices the basket is worth -0.20u. That's the overround, and it's what every buyer of this book pays. At the model's prices it's worth +0.72u, about 12% on the money. That second number is the whole bet, and it only exists if a unit conversion of BetMGM's win totals knows something Polymarket doesn't. The model has no football opinion. It converts one price into another and reports the gap.

So this is a test, run in public, with real money. Both baskets settle on February 14. Both go on the scorecard, and I'll refresh the numbers on the show as the season sorts out who was cheap and who was just bad.

Disclosure: Matt's and Andrew's baskets are real positions, listed on our scorecard. Our Polymarket links carry a referral code. The model basket is simulated. None of this is betting advice.