Public track record
Fantasy Scorecard
How our projections actually turned out. We publish it because a ranking is only worth trusting if it's checked. All scoring here is PPR.
Tested on five past seasons
The model as it runs today, replayed week by week over the 2021–2025 regular seasons using only what was known before each kickoff, then graded exactly the way the records below are. 31,044 projections across 85 weeks. How much each part of a projection counts was fitted on 2021–2023, so the numbers up top are 2024 and 2025 only: seasons it had never seen.
Average miss, 2024 and 2025
3.86 pts
against 4.23 for the no-model guess, on the same 12,419 projections
Order score
0.67
against 0.62 for the guess. 1.00 would mean our ranking matched the results exactly
Start calls that held up
54%
against 53% ranking the same players by the guess, with the same cut-offs
Inside our typical range
44%
of scores. It's meant to hold about half, so our ranges run a little narrow
Better than no model in 2024 and 2025, the seasons it had never seen, and in each of the five.
The no-model guess is each player's average so far that season, or last season's rate before his first game. It's a harder benchmark than it sounds: most of what any projection knows is how much a player gets the ball, and his average already carries that. In 2024 and 2025 we missed by 0.37 points a week less per player (about 9%), ranked players better, and got a few more start calls right. The edge held at every position. Over all five seasons, including the three it was fitted on, the miss was 3.92 against 4.33.
One week's fantasy score is mostly luck, so even a good projection misses by around 4 points on average. The margins that separate one method from another are small ones, and we'd rather show you ours than round it up.
Season by season
Ours / the guess's. Players with no designation on the official injury report (see below). Seasons marked “fitted” are the ones the model learned from, so they flatter it.
| Season | Projections | Avg miss | Order score | Start calls held |
|---|---|---|---|---|
| 2021 · fitted | 5,911 | 4.04 / 4.47 | 0.65 / 0.60 | 53% / 50% |
| 2022 · fitted | 5,911 | 3.90 / 4.36 | 0.65 / 0.60 | 54% / 51% |
| 2023 · fitted | 5,940 | 3.83 / 4.23 | 0.67 / 0.62 | 55% / 52% |
| 2024 | 5,982 | 3.87 / 4.20 | 0.67 / 0.63 | 55% / 53% |
| 2025 | 6,008 | 3.80 / 4.23 | 0.67 / 0.61 | 53% / 51% |
By position
Same players, 2024 and 2025 only.
| Pos | Projections | Avg miss | Order score | Start calls held |
|---|---|---|---|---|
| QB | 1,233 | 6.01 / 6.51 | 0.54 / 0.47 | 53% / 51% |
| RB | 2,965 | 3.88 / 4.18 | 0.77 / 0.74 | 67% / 66% |
| WR | 4,837 | 3.88 / 4.29 | 0.70 / 0.65 | 47% / 45% |
| TE | 2,955 | 2.80 / 3.17 | 0.67 / 0.63 | 48% / 44% |
What makes it a fair test. How much each part of a projection counts (this season against the past, work against points, the game's betting lines) was fitted on 2021–2023, so those seasons can't test it fairly; the numbers up top and the position table use 2024 and 2025 only. The rest of the model's settings weren't fitted on any of these seasons, with one exception: the discounts for players listed as questionable or doubtful were measured on 2021–2025. So the tables use only players with no designation on the official injury report, where those discounts play no part; over all five seasons their average miss is 3.89 against 4.30, the same picture as everyone together.
Where the replay differs from the live model. Past weeks have no game-day injury report, so the NFL's official Friday report stands in, and players on injured reserve (who aren't on that report) were sometimes projected anyway. That makes the replay harder on us than live use. In the other direction, the betting lines are at or near the close, and wind is what was recorded on the day.
Players who didn't play. Like the records below, these grade the games a player actually played. Counting a zero for every top-40 player who logged nothing (1,362 of them, many on injured reserve for the reason above), the average miss is 4.13 against 4.62 for the guess.
Backtest: this season so far
Weeks 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 re-projected using only what was known going into each week, then compared with what happened. It gives us a record from day one while the live one builds. It skips the injury report, since today's report reflects what happened in those games.
Starter calls that held up
56%
of 678 "Must start" and "Start" calls finished as starter-level
Bench calls avoided
79%
of 763 "Bench" calls finished outside starter level
Average miss
5.9 pts
1716 projections. We ran 1.5 low on average
Inside our typical range
61%
of scores landed inside the range we showed. It is meant to hold about half, the middle of likely outcomes
Against no model at all
The simplest thing you could do instead of reading us is take each player's average so far this season — or last season's, before he has played. Over the same 1,703 projections that guess missed by 6.5 points a week. We missed by 5.9, and we were closer on 58% of them.
Both numbers are judged on exactly the same players and weeks, using only what was knowable before kickoff. We benchmark against this rather than another site because it needs nobody else's data to check — you can reproduce it yourself from public stats.
Did each verdict mean something?
Average projected vs average actual points for everyone we put in each tier. The tiers should step down.
By position
Order score: 1.00 means our ranking matched the results exactly, 0 means no better than guessing.
| Pos | Players | Avg miss | Order score |
|---|---|---|---|
| QB | 286 | 6.4 | 0.40 |
| RB | 422 | 6.2 | 0.46 |
| WR | 688 | 5.9 | 0.42 |
| TE | 320 | 5.1 | 0.31 |
By week (average miss): Week 1 6.6 pts (182) · Week 2 5.5 pts (185) · Week 3 5.5 pts (180) · Week 4 6.2 pts (179) · Week 5 6.5 pts (172) · Week 6 5.6 pts (158) · Week 7 5.7 pts (158) · Week 8 6.3 pts (161) · Week 9 4.9 pts (172) · Week 10 6.0 pts (169)
Biggest busts (players we said to start)
- David Montgomery RB · HOU · Week 2 · called must startproj 23.7 → 4.4
- CeeDee Lamb WR · DAL · Week 5 · called must startproj 22.0 → 2.9
- Caleb Williams QB · CHI · Week 2 · called must startproj 25.5 → 7.7
- Tyler Shough QB · NO · Week 5 · called startproj 17.3 → -0.2
- Jaxson Dart QB · NYG · Week 2 · called startproj 18.1 → 0.8
Biggest booms (scored far above projection)
- Tetairoa McMillan WR · CAR · Week 4 · called startproj 11.6 → 45.2
- George Pickens WR · DAL · Week 10 · called startproj 12.6 → 43.5
- Davante Adams WR · LA · Week 2 · called flexproj 9.0 → 39.5
- Tyler Allgeier RB · ARI · Week 8 · called benchproj 5.9 → 36.1
- Emmett Johnson RB · KC · Week 10 · called benchproj 6.3 → 35.1
Not counted: 112 ranked players who didn't record a stat (inactive or unused), since we can't grade a score that never happened.
Live record, building
Every projection is frozen in our database before kickoff and graded after the game. Nothing is edited or added afterwards, so this is the real track record.
Freezing started part-way through this season, so this record is still small — and it spans a change we made to the projection on 25 September 2026, which moved the model faster when a player's role changes. Weeks before that date were produced by the older version. We are leaving them in rather than quietly dropping the ones that flatter us less. On 18 October 2026 we changed the starting point of every projection: it now begins from each player's work (the points his targets, carries and throws are usually worth) as well as his points. Projections frozen before that date come from the previous version, and they stay in too.
Graded so far: weeks 2, 3, 4, 5, 6, 7, 8, 9, 10. 597 more projections are waiting on games to finish.
Starter calls that held up
56%
of 567 "Must start" and "Start" calls finished as starter-level
Bench calls avoided
80%
of 702 "Bench" calls finished outside starter level
Average miss
5.7 pts
1504 projections. We ran 1.1 low on average
Inside our typical range
60%
of scores landed inside the range we showed. It is meant to hold about half, the middle of likely outcomes
Against no model at all
The simplest thing you could do instead of reading us is take each player's average so far this season — or last season's, before he has played. Over the same 1,490 projections that guess missed by 6.1 points a week. We missed by 5.7, and we were closer on 56% of them.
Both numbers are judged on exactly the same players and weeks, using only what was knowable before kickoff. We benchmark against this rather than another site because it needs nobody else's data to check — you can reproduce it yourself from public stats.
Did each verdict mean something?
Average projected vs average actual points for everyone we put in each tier. The tiers should step down.
By position
Order score: 1.00 means our ranking matched the results exactly, 0 means no better than guessing.
| Pos | Players | Avg miss | Order score |
|---|---|---|---|
| QB | 230 | 6.0 | 0.38 |
| RB | 388 | 5.9 | 0.54 |
| WR | 589 | 5.7 | 0.41 |
| TE | 297 | 5.0 | 0.34 |
By week (average miss): Week 2 6.7 pts (32) · Week 3 5.1 pts (259) · Week 4 6.1 pts (193) · Week 5 6.1 pts (183) · Week 6 5.3 pts (171) · Week 7 5.5 pts (163) · Week 8 6.1 pts (161) · Week 9 5.0 pts (174) · Week 10 6.0 pts (168)
Biggest busts (players we said to start)
- CeeDee Lamb WR · DAL · Week 5 · called must startproj 21.5 → 2.9
- Josh Jacobs RB · GB · Week 9 · called must startproj 18.1 → 0.5
- Zay Flowers WR · BAL · Week 8 · called must startproj 16.7 → 0.0
- Lamar Jackson QB · BAL · Week 8 · called startproj 16.5 → 0.0
- Drake Maye QB · NE · Week 10 · called must startproj 18.5 → 3.0
Biggest booms (scored far above projection)
- Tetairoa McMillan WR · CAR · Week 4 · called startproj 12.0 → 45.2
- George Pickens WR · DAL · Week 10 · called startproj 12.6 → 43.5
- Tyler Allgeier RB · ARI · Week 8 · called benchproj 6.0 → 36.1
- Tony Pollard RB · TEN · Week 7 · called flexproj 9.1 → 36.5
- CeeDee Lamb WR · DAL · Week 4 · called must startproj 15.3 → 41.3
How to read this
"Held up" means a player we called Must start or Start finished as a starter-level performer at his position that week: top 12 for quarterbacks and tight ends, top 24 for running backs and receivers. "Bench calls avoided" is the mirror image.
Fantasy scoring is noisy. A good projection is not one that always hits the exact number, it's one whose tiers step down the way ours should and whose range holds most scores. Only players who recorded a stat are graded, and we only grade PPR.
Stats from nflverse (CC-BY 4.0). Projections are estimates, not guarantees.