Zoltar is my NFL football prediction computer program. It uses a neural network and a type of reinforcement learning. The Super Bowl is coming up this weekend on Sunday, February 8, 2026. The Seattle Seahawks are playing the New England Patriots. For many years, a tradition for me is to go over to the huge house of my good friend Ken L to watch the game with about 20-30 other friends and family.

I always run the Super Bowl party contest where people guess the score at the end of each quarter, and at the end of the game. I create a 10-by-10 grid with rows numbered 0 to 9, and columns numbered 0 to 9. The rows represent the last digit of the score of one team (let’s say the Seahawks), and the columns represent the last digit of the score of the other team (say, the Patriots).

Click to enlarge

If there are 30 people at the party, every person gets to pick 3 cells (leaving 10 unused cells) and write their name in the 3 cells. Now, suppose the score at the end of the third quarter is Seahawks 23 – Patrioys 17. The person who owns cell (3,7) wins the third quarter prize. There are prizes for the score end of first quarter ($20), at half time ($40), end of third quarter ($40), and end of game ($100).

If nobody owns the cell that corresponds to a score, there are several ways to determine a winner. I use a spiral approach where I move one cell left, then up, then right, then down; then upper left, upper right, lower right, lower left, continuing until an occupied cell is found.

Now, at Ken’s Super Bowl party, there are usually only two math experts — Ken and me. Everybody else, except us two, just picks random cells in the game matrix. But Ken and I are well aware that some scores are much more common than others.

I did some analysis to look at most common scores. There were 272 games played during the 2024 regular season: there are 32 teams and each team played 17 games, so (32 * 17) / 2 = 272. And there were 12 playoff games, for a total of 284 games played so far in the 2025 season.

The six most common scores are:

Score Count
-----------
 20    42
 27    39
 24    35

 10    27
 17    26
 31    26

The least common scores are:

Score Count
-----------
 11     1
 47     1
 52     1

  1     0
  2     0
  4     0
  5     0

So, the nine best cells to pick in the game matrix are (0,0), (0,4), (0,7), (4,0), (4,4), (4,7), (7,0), (7,4), (7,7). But on the other hand, another good strategy is to pick cells not near anyone else.

In some years, we have a second Super Bowl party game where you have to predict the final score, for example, Seahawks 21 – Patriots 13. Based on the most common scores, and the fact that the Seahawks are roughly favored by 3 points, a good prediction for the 2025 Super Bowl is Seahawks over the Patriots 27-24, or 24-20.

It’s a super interesting problem to determine how to pick the final score best prediction. What if someone predicts very close scores, but an incorrect winner? What happens if one person is only off by 1 point for each of the two scores, but another person is off by 2 points on one score but gets the other score exactly right?

My system is named after the Zoltar fortune teller machine you can find in arcades. Fortune teller machines have been around for over 100 years. Here are three fortune tellers from the late 1800s made by the Roovers Brothers company (interestingly, the company is still in existence).

Right: “Jumbo” the elephant flips through a book of fortunes and stops at one for you to read. Interesting.