Fourth Down Lab.
One decision. Three ways to play it. Compare your chances of winning before judging what happened next.
Reading the defense…
Your decision probabilities will appear here.
How good does the play need to be?
Around the league
Loading the latest game snapshot…
| Situation | Bot says | Go / FG / Punt | Coach chose | Decision cost | |
|---|---|---|---|---|---|
| Loading fourth downs… | |||||
Where the numbers come from
This is a Data Dawgs browser port of nfl4th, the open-source model behind Ben Baldwin’s fourth-down bot, with nflfastR / fastrmodels win-probability components. The trained model artifacts were exported on September 28, 2026. It averages the possible outcomes of going, kicking, and punting; it does not judge a decision by whether one play worked.
The model uses situation, pregame spread and total, roof, clock, and timeouts. It does not know the play call, a particular kicker’s ability, current injuries, wind, or whether “one yard” is inches or a long yard. NFL regulation only; no overtime. Punt estimates are unavailable inside the opponent 31. Field goals of 71+ yards are excluded.
The default preserves the original bot’s 25-yard kickoff assumption for comparison. The 35-yard option is a scenario, not a forecast of the actual return. Late-game clock assumptions can change the call. This port also handles zero-time terminal states explicitly and applies user-entered runoff per outcome; those boundary cases may differ from upstream.
Weekly plays come from ESPN, with pregame lines from nflverse. This is a delayed snapshot, not a live betting feed; use the capture time. The refresh job is scheduled every 15 minutes during the NFL season, subject to GitHub Actions delays. Read the dated decision data, method, or open-source credits. Pup Port verified against published examples; not independently calibrated by Data Dawgs.