← Fourth Down / the technical article Archived prototype

// NFL forecast lab

Game on.
Let the model call it.

One featured matchup. A probability before kickoff.
A scorecard that keeps us honest afterward.

Featured matchup

Demo pairing · not this week’s schedule

Detroit

Detroit Lions logo

Lions

DET

Green Bay

Green Bay Packers logo

Packers

GB

Who gets the win?

Try the reveal with an illustrative forecast.
Sample values only. The trained model is not connected.

Correct / settled
0 / 0
Featured games only
Pick accuracy
—
No settled predictions yet
Probability quality
—
Brier score · lower is better

The first pick starts the story.

Once live, every featured forecast stays here: the pick, its probability, the final score, and whether we got it right. Wins and misses.

This record will cover the featured matchups, with comparisons against the same games’ baseline forecasts. Historical testing will be reported separately.

// the next chapter

From experiment
to game day.

Follow the build →
  1. 01

    Choose the week’s matchup

    A weekly schedule, team identities, and one featured game selected before kickoff.

  2. 02

    Save the model’s call

    Show the win probabilities and keep the original forecast, timestamp, and model version.

  3. 03

    Let the result speak

    Settle the game, keep the misses, and update both pick accuracy and probability scores.

A Thomas Howard project. Interface preview; live predictions are not available yet. Team logos via nflverse team metadata and ESPN. Independent of the NFL and its teams.