Model accuracy & track record
Every NFL model here, graded against real results.
How these are graded
Weekly projections: team environment x volume shares x efficiency x xTD allocation (DraftKings), graded per player week against the box score. Draft tiers: frozen preseason consensus, clustered by position, refit leave one season out and scored on the real season. Game model: QB adjusted Elo, graded against the market with the vig removed. Numbers refresh as results land.
This season, locked before kickoff
2026 live recordLive record
Picks lock at kickoff and grade from the final score. Nothing is added or removed after kickoff.
Weekly projections
2025 holdout, every playerEvery weekly projection, graded against the box score.
Weekly projections
r is the correlation between projected and actual DraftKings points on the 2025 holdout. Form is a recent form baseline, a higher r beats it. The rank bands below are measured on the 2025 holdout, a season the model was not fitted on. Live weeks grade as they finish. The headline stays on the 2025 holdout until the live sample is big enough.
r to actual · form = recent form baseline · 2025 holdout
// What a RB rank here has actually been worth
Where players we ranked in each band really finished, across the 2025 season. The bar is the middle half of outcomes, the notch is the median. Wide bars are the honest part: this is a weekly game and the order is a lean, not a promise.
Measured on the 2025 holdout, the season nothing was fitted on. Sample sizes run 54 to 432 player weeks per band.
// What a WR rank here has actually been worth
Where players we ranked in each band really finished, across the 2025 season. The bar is the middle half of outcomes, the notch is the median. Wide bars are the honest part: this is a weekly game and the order is a lean, not a promise.
Measured on the 2025 holdout, the season nothing was fitted on. Sample sizes run 54 to 432 player weeks per band.
// What a QB rank here has actually been worth
Where players we ranked in each band really finished, across the 2025 season. The bar is the middle half of outcomes, the notch is the median. Wide bars are the honest part: this is a weekly game and the order is a lean, not a promise.
Measured on the 2025 holdout, the season nothing was fitted on. Sample sizes run 54 to 192 player weeks per band.
// What a TE rank here has actually been worth
Where players we ranked in each band really finished, across the 2025 season. The bar is the middle half of outcomes, the notch is the median. Wide bars are the honest part: this is a weekly game and the order is a lean, not a promise.
Measured on the 2025 holdout, the season nothing was fitted on. Sample sizes run 54 to 432 player weeks per band.
// What a DST rank here has actually been worth
Where players we ranked in each band really finished, across the 2025 season. The bar is the middle half of outcomes, the notch is the median. Wide bars are the honest part: this is a weekly game and the order is a lean, not a promise.
Measured on the 2025 holdout, the season nothing was fitted on. Sample sizes run 54 to 108 player weeks per band.
// How this board has actually scored
Rank correlation with what players really scored, on the drafted universe, for every season we can grade. Two yardsticks: last year’s points, which is the naive answer, and the market’s average draft position, which is the hard one.
| Season | Players | Us | Last year | Market | First 50 picks |
|---|---|---|---|---|---|
| 2022 | 222 | 0.639 | 0.640 | 0.652 | 0.267 v 0.334 |
| 2023 | 327 | 0.703 | 0.655 | 0.762 | 0.317 v 0.224 |
| 2024 | 469 | 0.757 | 0.731 | 0.759 | 0.204 v 0.342 |
| 2025 | 468 | 0.768 | 0.763 | 0.760 | 0.275 v 0.506 |
Read that honestly. We beat last year’s points in 3/4 seasons, which is the bar. We beat the market in 1/4 — so this is market competitive, not market beating. And in the first fifty picks the market is clearly ahead. The pooled number is carried by the deep pool, where the only question is whether a player starts at all.
- Rookies are absent entirely: no prior history to project from, so this scores veterans only.
- ADP prices holdouts, camp battles, suspensions and trades that this model cannot see.
Draft tier track record
2019-2024How each draft tier actually finished, full PPR.
Draft tiers
Graded leave one season out, so a season’s tiers never see that season’s results. Median finish is the middle actual positional finish. League winner is an elite year, top 3 at QB, TE, K and DST, top 6 at RB and WR. Bust is a finish outside the startable range. A tier one rank wide has no tier record, one player held it. QB and TE top tiers are small samples. Kicker and defense are graded but not shown.
Running back
| Tier | Draft cost | Median finish | League-winner | Bust | Sample |
|---|---|---|---|---|---|
| T1 | RB1-6 | RB13 | 31% | 19% | 32 |
| T2 | RB5-15 | RB11 | 28% | 12% | 57 |
| T3 | RB15-24 | RB23 | 12% | 28% | 25 |
| T4 | RB16-32 | RB26 | 5% | 35% | 40 |
| T5 | RB19-41 | RB34 | 5% | 47% | 43 |
| T6 | RB24-50 | RB43 | 2% | 58% | 57 |
| T7 | RB33-62 | RB45 | 2% | 70% | 57 |
| T8 | RB42-68 | RB77 | 0% | 85% | 59 |
Wide receiver
| Tier | Draft cost | Median finish | League-winner | Bust | Sample |
|---|---|---|---|---|---|
| T1 | WR1-7 | WR9 | 45% | 16% | 31 |
| T2 | WR5-19 | WR14 | 27% | 18% | 33 |
| T3 | WR8-22 | WR20 | 9% | 18% | 55 |
| T4 | WR10-38 | WR38 | 3% | 41% | 74 |
| T5 | WR24-45 | WR37 | 6% | 38% | 52 |
| T6 | WR39-57 | WR60 | 2% | 64% | 58 |
| T7 | WR43-60 | WR52 | 3% | 64% | 39 |
| T8 | WR55-65 | WR77 | 0% | 84% | 38 |
Quarterback
| Tier | Draft cost | Median finish | League-winner | Bust | Sample |
|---|---|---|---|---|---|
| T1 | QB1-3 | QB7 | 20% | 10% | 10 |
| T2 | QB2-4 | QB4 | 43% | 0% | 7 |
| T3 | QB3-8 | QB9 | 37% | 21% | 19 |
| T4 | QB5-13 | QB16 | 11% | 52% | 27 |
| T5 | QB9-18 | QB14 | 6% | 39% | 31 |
| T6 | QB14-24 | QB22 | 3% | 69% | 32 |
| T7 | QB19-32 | QB26 | 0% | 71% | 31 |
| T8 | QB25-34 | QB27 | 0% | 82% | 22 |
Tight end
| Tier | Draft cost | Median finish | League-winner | Bust | Sample |
|---|---|---|---|---|---|
| T1 | TE1 | one player deep, no tier record | 6 | ||
| T2 | TE2-5 | TE6 | 36% | 36% | 14 |
| T3 | TE4-8 | TE11 | 7% | 43% | 14 |
| T4 | TE5-11 | TE13 | 15% | 35% | 20 |
| T5 | TE9-15 | TE13 | 5% | 45% | 20 |
| T6 | TE12-26 | TE22 | 0% | 74% | 31 |
| T7 | TE14-32 | TE23 | 5% | 67% | 39 |
| T8 | TE19-34 | TE29 | 0% | 83% | 30 |
Game & win model
2019-2025 holdoutPowers Winners, Survivor and Pick 6. Graded straight up on 1,865 games.
Game model
Brier scores the win probability, lower is better. The market moneyline is the ceiling. The blend matches it on priced weeks and the model alone covers unpriced weeks.