About & methodology

The project

The 2 Minute Drill Game is an independent fan project. It is free, has no accounts, and stores your progress only in your own browser. Fan-made. Not affiliated with or endorsed by the NFL, its teams, players, or the NFLPA. Team and player names are used only to identify real games.

Situations

Every drive comes from a real NFL game since 2016: fourth quarter, five minutes or less, one-score margin, normal offensive play. We take one starting point per real drive, skip spots where the leader could simply kneel out the clock, and skip games that were already decided (below 5% or above 95% win probability). That leaves 3,573 situations. The daily drive uses the most recent seasons (2022 onward) and mixes trailing, tied and leading spots about 4:2:1 across a week.

Players shown are the quarterback, two running backs, three receivers and a tight end who were most involved for that team in that game. Their numbers (catch rate, yards per target, interception rate) are season-to-date through that week. Each receiver also gets a 40–99 rating: yards and catch rate per target compared with others at his position (counted less when he has few targets), interceptions when targeted, and his targets per game. The rating is a summary for the receiver cards; the engine uses the underlying numbers.

The engine

Play outcomes

Every play since 2016 is classified into one of 27 calls (for example a quick pass to the strong side, or an inside run to the weak side) using nflverse play-by-play and FTN charting for play-action, RPOs, screens, rollouts, sneaks, trick plays, box counts and blitzes. For each call, down, distance bucket, field zone and defensive look we keep a sample of real late-half outcomes — yards, completion, sack, interception, fumble, out of bounds — falling back to broader groups when a bucket has fewer than 80 plays. Play-action is rare late in games, so its buckets blend its own outcomes (same field zone, look and down type) with regular dropbacks in the same situation, weighted by how much play-action data there is. When you call a play the engine draws one real outcome from the matching bucket. A touchdown drawn from a bucket still scores from a little farther out, with a chance that falls off with the extra yards needed. Penalties are drawn at their real late-game rates. Under each result the game shows the real play it was drawn from (season, teams, clock, down and the official play description). When your receiver, an out-of-bounds try or a longer touchdown run changed that outcome, it says “Adjusted from a real play”.

Team strength

The closing point spread sets how much better one team is: on each snap the favorite sometimes gets the better of two drawn outcomes and the underdog the worse, sized so one point of spread moves late-game win chance about as much as it does in nflverse's Vegas-adjusted model (about 0.6 points of win chance). A trailing offense also gets a small edge, which stands in for prevent defenses and for real play-callers checking into what the defense gives them; without it, simulated comebacks fell short of real ones.

Clock

How long a play takes and how much time runs off before the next snap are measured from consecutive snaps in late-game drives, by tempo: no-huddle, normal huddle, or a leading team using the play clock. Rules are layered on top: incompletions and plays out of bounds stop the clock in the last five minutes, the two-minute warning, timeouts, spikes and the 40-second play clock.

Overtime

A tie at the end of regulation goes to overtime under the current rules: 10 minutes in the regular season (15 in the playoffs), both teams get the ball unless the first possession ends in a defensive score, then the next score wins. Regular-season games can still end tied. Finish level in any mode and you call the overtime yourself, with win chances from rollouts under the same overtime rules (overtime calls are not graded). While grading regulation calls, a tie is worth that game's overtime win chance, so the stronger team is not treated as a coin flip.

Kicking

Field goal probability is a logistic model of distance fit to every attempt since 2021, adjusted for the actual kicker's accuracy relative to expectation (shrunk toward average). Extra points, punts, kickoffs (2025 rules: touchbacks to the 35) and onside kicks use their recent empirical distributions.

Win probability and grading

From any moment, the engine plays out the rest of the game hundreds of times with both teams following a baseline NFL policy: run/pass mix, fourth-down and two-point decisions, tempo, timeouts and onside kicks by score and time, measured from real coaches. Each option you could choose is evaluated with the same random numbers (common random numbers), then the best few and your choice get a second, larger batch to avoid rewarding lucky estimates. Grades: within 0.5 points of the best option is green, 2 points yellow, 5 points orange, worse is red.

Percentile and “Coached like”

Your total cost is ranked against 4,000 simulated play-callers whose choices follow NFL frequencies with varying skill. The pool is tuned so real NFL coaches, graded on their own drives the same way, land in the middle of it (a median around 50). “Coached like” compares four late-game tendencies — pass rate, deep-shot rate, fourth-down aggressiveness and two-point rate, each relative to the league in the same situation — with 64 NFL head coaches who have enough late-game snaps since 2016.

Calibration

On 600 held-out 2024–2025 situations, the engine's starting win probability scored a Brier score of 0.160 against actual results, a little better than nflfastR's win probability model (0.166). Trailing teams average 30% in the engine where real teams won 29%, and leading teams 85% where real teams won 85%. The team-strength and two-minute settings were fit on 2016–2023 games, so these 2024–2025 numbers are an honest check rather than a fit.

Limitations

• Team strength comes from the closing point spread, plus your receivers and kicker. The game doesn't know about injuries, matchups or game plans.
• Strong/weak side splits come straight from the data and can carry sampling noise.
• The field shows a schematic 11-on-11 picture: a standard formation, routes for the call and the ball going where the result says. It is not the real players' positions.

Data and credits

Play-by-play, schedules, rosters and players from nflverse; charting from FTN Data via nflverse. Licensed CC BY-SA 4.0. Data current through the 2026 season, week 6(built 2026-10-17). Team colors are used only beside abbreviations; no logos or player images are used.

Questions or corrections? Read the guide or the privacy policy.