Omaha 2.4.0
Released 9 October 2026
The simulation now knows whether the quarterback is in the shotgun or under center on every snap. None of the eleven figures below moved: the average is +10.0% before and after, and what changed is how a game plays out rather than how it scores.
How to read these numbers
Every figure below compares the simulation against a player’s own season average so far, scaled to how much of the game is left, roughly what a reader could work out without us. Zero means we add nothing over that. Positive means we beat it; negative means we are worse than it.
Measured across the 2024 season, weeks 1–18, rebuilding the model each week using only what was known at the time. 400 simulations per situation.
The change column is simply this version’s accuracy minus 2.3.0’s, each measured across everything that version covers, so the two columns you can see always account for it. We also measure the change on shared situations alone, which is a more sensitive test; where the two disagree we publish this one.
Accuracy by stat
| Stat | vs. season average | Change from 2.3.0 |
|---|---|---|
| Receiving touchdownsrare event | +20.1% | +0.14 |
| Rushing touchdownsrare event | +16.7% | +0.01−0.37 to +0.40 |
| Receiving yards | +14.4% | −0.06−0.09 to −0.00 |
| Pass attempts | +13.0% | −0.01−0.15 to +0.12 |
| Passing yards | +11.0% | −0.10−0.28 to +0.08 |
| Rushing yards | +10.9% | +0.04−0.01 to +0.13 |
| Completions | +9.8% | −0.09−0.28 to +0.09 |
| Passing touchdownsrare event | +7.2% | +0.15−0.17 to +0.76 |
| Receptions | +3.3% | −0.00 |
| Targets | +3.2% | +0.01−0.07 to +0.10 |
| Rush attempts | +0.9% | +0.18+0.07 to +0.27 |
| Average across all figures | +10.0% |
The average is an unweighted mean across the figures above. It is a summary of how we are doing, not a single score for the model: yards and counts are not measured in the same units, so combining them any more cleverly than this would just let passing yards decide the answer.
Each change carries the range the measurement can actually support. A change whose range crosses zero is shown in neutral rather than as a gain: we cannot tell it apart from no change, and we are not claiming it as one.
What changed
Where the quarterback lines up changes who gets the ball. From the shotgun he keeps it himself far more often, and his tight end is thrown to less. Under center the opposite holds. The simulation had no idea which one a snap was. It now decides on every play, from the down and distance, the score, the clock, the team's own habits and how long its quarterback has been in the league.
Across 2022 to 2025, quarterbacks take 7% of designed runs from under center and 20% from the shotgun, and tight ends 29% of targets against 21%. The simulation's shotgun rate tracks real teams by situation: 90% when trailing by two scores in the fourth quarter (real 87%), 47% when leading by two scores (real 41%), and 98% on third and long (real 98%).
A team chasing the game goes to the shotgun and its quarterback runs more; a team protecting a lead goes under center and he runs less. Both now happen, in the right direction, though not yet by enough.
Trailing by two scores in the fourth quarter, quarterbacks take 21% of real runs. The simulation gave them 12% and now gives 15%. Leading by two scores, real is 5%: the simulation gave 14% and now gives 12%. Measured on 2025 snaps the simulation had not been trained on.
Where a quarterback lines up decides who gets the ball on a given snap, and over a whole game those choices largely balance out, so a player's total moves very little. We built it so that a team's usual mix of formations changes nothing on average: the simulation already knew a team's habits from its own plays, and counting them twice would have been a mistake. What changes is the game in between: which plays happen when the score and clock make one formation likelier than the other.
Rush attempts +0.7% to +0.9%, the only change that holds up on a game-by-game resampling, and receiving yards 0.04 points lower. Every other figure moved by less than two tenths of a point, inside what a change that does nothing moves.
What this version still gets wrong
- Quarterback runs still barely respond to the score. Real teams range from 5% of runs to the quarterback when protecting a lead to 21% when chasing; the simulation now ranges from 12% to 15%. Most of the rest is very likely scrambles, a called pass that becomes a run, which the simulation does not model as its own event yet.
- The simulation puts teams in the shotgun a few points more often than they are this season. The league has been moving away from it: 70% of snaps in 2024 against 66% in 2025, and still falling, while the model learns mostly from the earlier season.
- Tight ends are still thrown to too often, about five points above real across every situation. Where the quarterback lines up does not fix that; it is a separate problem.
- None of this reaches the figures on this page, and we are saying so rather than finding a number to lead with. It shipped because the simulation now behaves more like real football, and it costs nothing measurable to get there.