Skip to the page

Meet Newton

Newton is Saturview’s forecasting model. It is software, not a person: it reads college football data, saves a pick for each rated game before kickoff and is graded on it in the open.

What Newton does

  • Picks winners. It makes a pick for each rated FBS matchup in the next seven days, with a win chance and a projected score.
  • Reads pregame information. The current model, Newton v3, forecasts a game’s margin and win chance from team strength, how each offense and defense has played this season against the opponents it faced, rest and travel. It was built on earlier seasons and tested on full seasons it had not seen.
  • Updates until kickoff. A pick can be remade as new information arrives. A confirmed absence of a team’s starting quarterback can change a pick.
  • Answers questions. Ask Newton answers from Saturview’s saved records about teams, players, games and Newton’s own picks. It explains a pick and never changes one.

What Newton does not do

  • It does not read betting lines. In our tests it did not beat the betting line either, and it gives no betting advice.
  • It does not change a pick after kickoff.
  • It does not measure injuries or weather, apart from a confirmed starting quarterback absence for that one game.
  • It does not promise results. Projected scores and win chances are estimates.
  • It is not trained on your personal information or your Ask Newton questions.
  • It can make mistakes when you ask it questions, so treat its answers as a starting point for the football conversation.

How its record is kept

  1. Saved. A pick is made for each rated game in the next seven days.
  2. Remade. When new information changes it, the new version is saved. Earlier versions are kept.
  3. Locked. At kickoff the pick locks and nothing about it changes.
  4. Graded. The final score decides. The last version made before kickoff is graded, never an earlier one and never anything made afterwards.
The life of one pick.

Every pick counts, including games that are not featured on the front page. Postponed or canceled games are not counted in the correct-pick percentage, and a corrected final score updates the result while the pick stays as it was.

The season record has two parts, and both are labeled wherever they appear. The live record counts only picks made before kickoff. The season number adds a replay of weeks 1 and 2, worked out after those games were played, so it always says it includes the replay. Each pick stays under the model that made it, and a newer model is never scored on games an older model already picked.

The season record is on the front page. A free account can read every settled pick. Upcoming picks are part of Premium.

How accurate it has been

In replays of the 2024 and 2025 seasons, 1,498 games, with each season predicted using only the seasons before it, Newton v3 picked 72.0% of winners. The original rating model picked 71.7%. The gain is modest, and a replay is not a live record.

The write-up, with every step and every idea that was dropped, is in the research notes: How Newton v3 was built. How each source and model is used is on the Methodology page.

Why Newton writes in the first person

Newton says “I had” and “Called it”, not “Newton’s pick was right”. A pick is a claim, and the sentence should say who is making it. That is why a miss reads “Missed this one” and is said out loud rather than buried. It is still a model. If you ask, it will tell you it is an AI, and it never claims to be a person.

Labels and page titles keep the name, as in “Newton’s record”. A sentence Newton says about its own pick uses “I”.

Keep reading