How it works

Every game on FlexPicks gets two predictions: who wins, and what the score might be. They come from two different models, and knowing that makes the cards a lot easier to read.

The win model picks the winner

The first model answers one question: how likely is the home team to win? It looks at how both teams have played in recent weeks and turns that into a percentage. The team with the better chance gets the FlexPicks badge, and the bar on each card shows that chance.

The score model projects the points

The second model is built separately and predicts how many points each team scores. Those are the gold numbers on each card. It's useful for seeing whether a game looks like a shootout or a grind, and how big the gap might be.

Why they sometimes disagree

Because the two models learn on their own, they can lean different ways when a game is close. If the win model picks one team but the score model has the other team scoring more, the card is marked as a split call. Treat those as genuine toss-ups.

Where the numbers come from

Both models are trained on years of public NFL play-by-play data from the nflverse project. Each team's recent form is summarized from that data, using only games that were played before the one being predicted.

When it updates

After Monday Night Football each week, the latest results feed back in, both models retrain, and new picks go up for the next week. The track record page grades every pick once the games are final.

The fine print

This is a hobby project for fun and for fantasy and pick'em leagues. It is not betting advice, and even a 75% pick loses one time in four.