Results as of the Week 5 games. This grades the pre-game board as we rebuilt it after the games, a sample of the format from before snapshots went up ahead of kickoff. From Week 6 on, each snapshot goes up first and gets graded as published.
The pre-game post showed what the data looked like. This one grades it: projections against real DraftKings points, the hits and misses, and whether the model earned its place on the board. One week is a small sample, so this page shows the method and the numbers and declares no winner.
The short version
- On the 45-player board, the model’s projections lined up worse with real results than the public average did: correlation 0.28 vs 0.54, average miss 9.1 vs 8.1 points. The model was closer than the public average on 17 of 45 players.
- Across all 267 graded players (not just the board), the model won on one measure and lost on the other: the model’s average miss was 4.92 points vs 5.51 for the 4 public sources (their individual scores averaged), with correlation 0.601 vs 0.618.
- The board is a selected group, the top-scoring names, so its correlations run lower than a full slate’s. Both views are shown for that reason.
- The largest model miss on the board was Chris Olave (model 24.3, actual 8.6). The largest beat was Isaiah Likely (model 5.4, actual 32.4).
Board results by group
| Group | Players | Avg model | Avg public | Avg actual |
|---|---|---|---|---|
| Sources split | 9 | 17.0 | 15.6 | 14.1 |
| Sources agreed | 36 | 13.6 | 14.6 | 18.2 |
Players where the sources agreed averaged 18.2 points; the split group averaged 14.1. Over the full 198-player pool, split players averaged 11.0 vs 9.2 for the rest. The slate-wide numbers point the other way. With only 9 players in the split group this week there is no reliable conclusion either way. We will keep tracking it.
Biggest misses and beats against the model
The largest gaps between the model’s projection and what happened, on the board.
Model too high
| Player | Pos | Team | Salary | Model | Public avg | Actual |
|---|---|---|---|---|---|---|
| Chris Olave | WR | NO | $7,700 | 24.3 | 17.4 | 8.6 |
| RJ Harvey | RB | DEN | $5,200 | 14.2 | 12.5 | 0.7 |
| Chase Brown | RB | CIN | $6,900 | 20.8 | 17.6 | 8.0 |
| Harold Fannin Jr. | TE | CLE | $4,400 | 13.0 | 10.5 | 1.4 |
| Nico Collins | WR | HOU | $7,500 | 21.2 | 17.1 | 10.6 |
| Michael Wilson | WR | ARI | $6,200 | 17.4 | 16.9 | 7.3 |
Model too low
| Player | Pos | Team | Salary | Model | Public avg | Actual |
|---|---|---|---|---|---|---|
| Isaiah Likely | TE | NYG | $4,300 | 5.4 | 12.1 | 32.4 |
| Jaylen Warren | RB | PIT | $6,700 | 15.2 | 16.9 | 39.8 |
| Drake Maye | QB | NE | $6,300 | 19.4 | 20.4 | 42.5 |
| Mike Evans | WR | SF | $6,500 | 9.8 | 14.1 | 29.3 |
| Jayden Daniels | QB | WAS | $6,000 | 16.4 | 19.2 | 35.3 |
| Braelon Allen | RB | NYJ | $5,400 | 4.7 | 13.8 | 21.6 |
Both lists are ordinary single-week variance, which is why the grade runs over many weeks.
Outside opinion
Players flagged bullish by logged articles and podcasts averaged 13.8 actual points across the 198 graded, versus 6.9 for neutral and 5.8 for bearish, the order you’d hope for. One week is not a trend. We will keep tracking it.
Injury tags
8 players on the board carried an injury tag at snapshot time and averaged 18.8 points against a model average of 11.7. The model discounted them, and as a group they beat that discount. With 8 players that is an observation to track, not a rule.
What this means
- The model did not beat the public average on this board. We will check again next week and look at where its big gaps came from.
- Across the full slate the result was mixed: the model won on one accuracy measure and lost on the other. Whether that holds over more weeks is the test.
- Publishing the snapshot first and the grade afterward means the grade cannot be adjusted to fit. Bad weeks are as useful as good ones.
Next week’s snapshot goes up before the games and gets graded the same way. The fuller discussion happens in the community.
FAQ
What is this page?
The grade for the Week 5 signals post: the pre-game projections compared with real DraftKings points, with the hits and the misses.
Did the projections change after the games?
Week 5’s board was rebuilt after the games, not published beforehand, so this week is a sample of the format. From Week 6 on, the grade is of a snapshot published before kickoff.
Why show both the board and the full slate?
The board is the top-scoring names, a selected group, so its correlations run lower than a full slate’s. The full-slate numbers cover everyone. The board covers the players the signals page featured.
Is one week enough to judge the model?
No. A single slate is noisy, so the same snapshot and grade are published every week.