NFL Game Simulator

How this NFL game simulator works

Most NFL simulators roll dice on team ratings. This NFL game simulator starts from what actually happened on the field instead. Every snap is simulated in the same loop: a real situation, your call, a real outcome drawn from comparable plays, the clock running the way it does in real games, and then hundreds of simulated finishes to see what your call was worth.

• Real situations. About 3,573 fourth-quarter spots from NFL games since 2016: five minutes or less, one-score games, with the real score, clock, down, distance, timeouts and players.
• Real outcomes. Every NFL play since 2016 is sorted into one of 27 calls. When you call a quick pass to the strong side on 3rd-and-6 from your own 40 against a blitz, the result is drawn from real late-half plays that match — falling back to broader groups when a bucket has fewer than 80 plays.
• A real clock. Time between snaps is measured from late-game drives: about 18 seconds in a hurry-up, 31 with a huddle and 35 when a leading team milks the play clock. The two-minute warning, timeouts and spikes work as they do on Sundays.
• Team strength. The closing point spread decides how much better one team is, so a big favorite gets slightly better outcomes on each snap.

What kind of NFL simulator is this?

"NFL simulator" covers very different tools, and it's worth picking the right one. Here is how they compare and where this one fits:

If you want to know who would win Eagles vs Cowboys in Week 10, a matchup simulator is the better tool. This NFL game simulator answers a different question: what should the coach call right now? If you've ever yelled "why didn't they spike it?" at the TV, this is the one for you.

Two playbooks, three modes

Rookie puts six calls on big buttons — run, short pass, deep shot, spike, timeout and field goal — so you can play with no football background. Pro opens all 27 plays, receiver targets with their real season numbers, tempo and every clock option. On defense you pick the look before each snap, and after a score you choose the extra point or a two-point try and the kickoff.

• Daily drive: one real ending a day, the same for everyone, with a streak and a daily rank.
• Practice: unlimited drives, filtered to trailing, leading or tied — the mode embedded above.
• Survivor: win real fourth quarters in a row until you lose one; ties go to an overtime you call.

How your calls are graded

Before the result of a play is shown, the engine plays out the rest of the game about 300 times for every option you could have picked, using the same random numbers for each so the comparison is fair, then runs a larger batch on the best few. Your grade is the gap between your call and the best one, in points of win probability: within half a point is the best call, more than five is a big miss.

At the end of a drive your total is ranked against 4,000 simulated NFL play-callers, and the real coach's calls on the same drive are graded the same way, so you can see whether you actually out-coached him.

A clock management simulator, too

Most of what separates good and bad late-game coaching is the clock, and that is what this simulator punishes and rewards most directly. Burning a timeout at 2:10, letting 25 seconds run before a spike or kicking a field goal with too much time left all show up in your grade. If you want the rules behind those decisions, start with how many timeouts each team gets and how a two-minute drill works.

How accurate is this NFL game simulator?

The engine's team-strength and late-game settings were fit on 2016–2023 games, then checked on 600 situations from 2024 and 2025 it had never seen. Its starting win probabilities scored a Brier score of 0.160 against the real results, a little better than nflfastR's widely used model (0.166). Trailing teams averaged a 30% win chance in the engine where real teams won 29% of the time; leading teams averaged 85% and won 85%.

It has limits, and we'd rather you know them: team strength comes from the point spread plus your receivers and kicker, so the simulator doesn't know about injuries or specific matchups; the field shows a schematic formation rather than real player positions; and strong- and weak-side splits carry some sampling noise. The full method is on the methodology page.