Zoltar is my NFL football prediction computer program. It uses a type of reinforcement learning and a neural network. The first game of the 2026 NFL season will be played on Wednesday, September 9, with the New England Patriots at the Seattle Seahawks, in a rematch of last year’s Super Bowl (won by the Seahawks 29-13).

To get ready for the upcoming season, I thought I’d do an analysis of last season’s results. The big issue with Zoltar (or any betting system on the NFL) is dealing with Vegas opening point spreads versus Vegas closing point spreads.

Up until about 2020, point spreads were relatively stable, meaning the opening point spreads on Tuesday morning (following the Monday night game) would not change much until the closing spreads on the day of the game — except in the case of a key player injury. A swing of more than 1.0 points was unusual.

But as the amount of money being bet on NFL games has increased to gigantic amounts — billions of dollars each week now — the point spread data now shows wild swings.

The opening point spread values are set by a handful of experts. But hundreds of millions of dollars can be instantly bet on a particular team on Tuesday morning. Sports books only make money if the amount bet on each pair of teams in a matchup is close to the same. So if a huge amount of money is quickly bet on team A based on the opening point spreads on Tuesday morning, the books must quickly and dramatically change the point spread to encourage betting on team B.

A swing of 7 or more points between Tuesday morning and Saturday afternoon is now common.

Anyway, I tracked down the historical closing point spread data for the 2025 season. I found the data at covers.com/sportsoddshistory/nfl-game-season/?y=2025. I massaged the data to get it into the format used by Zoltar and ran an analysis:

Analysis of 2025 season using Vegas closing odds

Betting threshold = 3.5
week =  1 | 3 2 0
week =  2 | 1 1 0
week =  3 | 3 0 0
week =  4 | 3 2 0
week =  5 | 1 3 0

Betting threshold = 4.5
week =  6 | 2 1 0
week =  7 | 2 1 0
week =  8 | 1 0 0
week =  9 | 0 2 0
week = 10 | 3 3 0
week = 11 | 0 0 0
week = 12 | 2 0 0
week = 13 | 1 1 0
week = 14 | 1 0 0
week = 15 | 4 0 0

Win Loss Push =  27 16 0
Accuracy = 62.8%

For weeks 1-5, I used a betting threshold of 3.5 points, meaning Zoltar would place a virtual bet on a team only if the difference of opinion between Zoltar’s prediction and the Vegas point spread is strictly more than 3.5 points. For example, if Vegas thinks team A will beat team B by 4 points (“team A -4”), Zoltar will place a bet on the favored team A only if Zoltar believes team A will win by more than 4 + 3.5 = 7.5 points.

For weeks 6-15, I used a more conservative threshold of 4.5 points. The idea here is that human bettors start getting more accurate after week 5.

I don’t analyze week 16 or 17 when injuries play a huge role. And I don’t analyze week 18, the last week of the regular season, when teams that have nothing to gain (already in the playoffs or already eliminated) will not play their best players to avoid injury.

For the 2025 season, using the Vegas closing point spreads, Zoltar scored 27 correct, 16 wrong (with no pushes) for 62.8% accuracy.

Good fun.

Las Vegas clearly sees sports betting as an important part of the future. It’s already a multi-billion dollar industry and it will soon reach multi-trillion dollar levels (actually, it may be there already).

Right: This is the showroom of the old “New Frontier” hotel and casino. The property was the “Last Frontier” from 1942-1955, the “New Frontier” from 1955-1967, and “The Frontier” from 1967-2007. The “Trump International” hotel is now on the site. In the photo, I love the mid-century modern atom age clock on the wall. I visited the “New Frontier” with my parents in 1968, and “The Frontier” many times during my blackjack card counting days with my good friend and UCI classmate Ed Koolish.