THE SHORT VERSION
How expected
points work
Expected points ask a simple question: what does this kind of workload usually produce?
A carry near the goal line has a better chance to score than a carry at midfield. A deep target offers a different opportunity from a short pass. We give those opportunities a value, then add them up for each player.
01 / THE OPPORTUNITY
One carry. Where does it happen?
Move the ball to see how the model values a running back’s carry.
for this carry
Carries inside the 5 get the most weight in this model.
These are historical averages fitted by field-position range. They don’t mean every carry from the 3 will score. The steps in the graphic reflect the model’s 20-, 10- and 5-yard boundaries.
02 / THE RESULT
Compare the workload with the score
Here’s Tyler Shough in Week 1 of 2026. His 56 pass attempts and 4 carries added up to a valuable workload.
A gap can come from skill, randomness or things the model misses. It doesn’t automatically make a player a buy or a sell.
03 / HOW TO USE IT
Look for changes across several games
Start with your position and league scoring. Open a player to see his weekly results and the carries, targets or passes behind the number.
- More expected points? Check whether the player is getting more opportunities or more valuable ones.
- A big actual score? See whether the workload grew too. One long touchdown can make a small workload look impressive.
- Choosing next week’s lineup? Check injuries, role changes and the matchup separately. They aren’t included here.
Choose 4 or 6 points per passing touchdown on the QB tab. Both expected and actual points update, and the order can change. Passing and rushing opportunities are valued separately.
After Week 1, the season average is just one game. “Recent usage average” blends the newest game with earlier usage, including last season at the start. It can take time to catch up with a new role.
THE EVIDENCE
How does it compare?
We checked six 2024 and 2025 tables from F4NT4SYF00TB4LLF4N, whose opportunity scores are hosted by Subvertadown.
The methods largely agree. In the 2025 exports, his receiver scores describe completed games more closely. His published method uses yards before catch; ours includes the depth of incomplete targets too.
Does recent usage help with the following week?
We averaged up to four earlier games from each table and compared that average with the next week’s actual score. Here is the average error in points for the 2025 matches. Lower is better.
We’re keeping the current usage model. A simpler model and a version using completed air yards did not consistently improve next-week results.
See the comparison details
Our 2024 test fits prices on 2021–2023. Our 2025 test fits on 2021–2024. Each prediction uses only earlier matched games. This tests a simple trailing average, not the site’s 35% weighted average.
| Year / position | Next-week cases | Our error | Other error | Difference: 95% interval |
|---|---|---|---|---|
| 2024 RB | 997 | 4.82 | 4.85 | -0.065 to 0.009 |
| 2024 WR | 1497 | 4.67 | 4.71 | -0.095 to 0.005 |
| 2024 TE | 549 | 3.51 | 3.56 | -0.096 to -0.008 |
| 2025 RB | 815 | 4.89 | 5.02 | -0.186 to -0.075 |
| 2025 WR | 1217 | 4.64 | 4.75 | -0.205 to -0.001 |
| 2025 TE | 613 | 3.69 | 3.72 | -0.095 to 0.045 |
Difference is our error minus the other score’s error. Intervals resample whole players 1,000 times. Three of six intervals include zero. These are exploratory comparisons without a multiple-comparison adjustment.
The exports cover different portions of each season and selected players. We retain zero scores when a matching usage row exists. Byes, games without an eligible usage row and unmatched names or positions are excluded. Neither absent-player risk nor the other author’s original publication timing is tested.
Full comparison and reproducible checksThe model, scoring and fitted values
How the estimate is calculated
We fit a separate linear regression for each position and scoring option on 2021–2025 regular-season nflverse player-games. The inputs are counts of carries, targets and passes, field-position ranges and air yards. The target is that game’s fantasy score.
Expected points = a per-game baseline + the value of each recorded opportunity. Red-zone terms stack: a carry inside the 5 also counts as a carry inside the 10 and 20. Negative game estimates are floored at zero.
These inputs are observed during completed games. They do not use the player’s realized touchdowns or yards after catch as features. Sacks, kneels and spikes are excluded from opportunity counts. The model can miss persistent player differences.
Scoring
Passing yards: 0.04 points. Passing TDs: 4 or 6. Interceptions: −2. Rushing and receiving yards: 0.1. Rushing and receiving TDs: 6. Two-point conversions: 2. Lost fumbles: −2. Receptions: 0, 0.5 or 1 for Standard, Half PPR or PPR. League bonuses are not included.
Six-point passing touchdowns use a separate fit to the corresponding actual score. The extra expected points are learned from usage, rather than added from the player’s actual touchdown count. The setting applies to passing touchdowns at every position.
Recent usage average
After each recorded game: 35% of that game’s expected points + 65% of the previous average. The starting average comes from last season, blended toward a position baseline when fewer than four games were recorded. Players without prior data start at that baseline. After four games, about 18% of the starting value remains.
Half-PPR coefficients
| Feature | QB 4pt | QB 6pt | RB | WR | TE |
|---|---|---|---|---|---|
| Per-game baseline | 0.96 | 1.37 | -0.36 | -0.12 | 0.08 |
| Pass attempt | 0.13 | 0.13 | — | — | — |
| Pass inside the 20 | 0.47 | 0.62 | — | — | — |
| Pass inside the 10 | 0.66 | 0.92 | — | — | — |
| Passing air yard | 0.02 | 0.02 | — | — | — |
| Carry | 0.60 | 0.59 | 0.55 | 0.89 | 1.15 |
| Carry inside the 20 | 0.59 | 0.66 | 0.30 | 0.54 | — |
| Carry inside the 10 | 0.77 | 0.63 | 0.37 | 1.24 | — |
| Carry inside the 5 | 1.16 | 1.08 | 0.84 | — | — |
| Target | — | — | 0.91 | 0.98 | 0.93 |
| Target inside the 20 | — | — | 0.24 | 0.25 | 0.65 |
| Target inside the 10 | — | — | 1.43 | 0.86 | 0.87 |
| End-zone target | — | — | 0.77 | 0.76 | 0.97 |
| Target 20+ yards deep | — | — | 1.06 | 0.03 | -0.12 |
| Target air yard | — | — | 0.02 | 0.03 | 0.03 |
Rounded here for readability. All coefficients and scoring options are included in each edition’s JSON download.
How well does usage explain a completed game?
We fit on 2021–2024 and evaluated on 2025, which the fit never saw. This uses the actual workload from the game being scored. It is a descriptive check, not a pregame forecast.
| Position / scoring | Games | Average absolute error | R² |
|---|---|---|---|
| QB (4pt TD) | 640 | 4.77 points | 0.544 |
| QB (6pt TD) | 640 | 6.05 points | 0.505 |
| RB | 1358 | 3.08 points | 0.622 |
| WR | 2134 | 2.95 points | 0.569 |
| TE | 1129 | 2.23 points | 0.609 |
R² describes how much score variation the model explains relative to a constant average. It is not an accuracy percentage. These results use half PPR. The live model then fits all five completed seasons for use in 2026.