A modest disagreement of 1.6 points.
Across five seasons, cards this close to the market claimed 54% accuracy and delivered 50%. The claim runs a little ahead of the delivery.
How much weight this card's information deserves — not a recommendation.
Why
what the model is reading-
Defense, in points
PSU +8.7 pts
- Pace and possessions 106 plays · 35th percentile
-
Home advantage
PSU +2.6 pts
Be careful about
- One book is quoting this game, so the market number has no depth behind it and may move sharply.
- Atlas and the market have moved opposite ways on the total since it opened.
Projection built Oct 29, 2026 4:14 AM ETMarket updated Oct 29, 2026 4:14 AM ETData status
New to Atlas? What a card is, and what a grade means
Market detail where the line opened, where it is now, and the prices
| Spread | Open | Current | Move |
|---|---|---|---|
| Penn State | −8.5 | −9.5 | +1.0 |
| Purdue | +8.5 | +9.5 | −1.0 |
| Price | −110 | −110 | — |
The spread has moved 1.0 toward Penn State and the total has moved 2.0 up since the open.
| Total | Open | Current | Move |
|---|---|---|---|
| Game total | 52.5 | 54.5 | +2.0 |
| Price (over) | −110 | −110 | — |
DraftKings · 1 book quoting
Projection detail the projected score, the chance of winning, the total's range and the most likely exact score
Atlas difference
| Measure | Atlas model | Market | Difference |
|---|---|---|---|
| Spread | PSU −11.1 | PSU −9.5 | +1.6 |
| Total | 54.3 | 54.5 | −0.2 |
| Penn State win probability | 76.5% | 74.1% | +2.4 pts |
Matchup how each offense compares with the defense it faces, season to date, ranked among 138 FBS teams
Season to date through 24 October: Purdue 7 games, Penn State 7 games. Ranks are among 138 FBS teams, 1st the best; the crest marks which side ranks better on each line.
edge Penn Stateedge Purdueedge Penn Stateedge Penn Stateedge Penn Stateedge Purdueedge Penn Stateedge Purdueedge Penn Stateedge Penn Stateedge Purdueedge Penn Stateedge Penn Stateedge Penn Stateedge Penn Stateedge Penn Stateedge Penn StateFigures come from completed games only, so every one was known before kickoff. They are raw, not adjusted for opponents - All drivers has Atlas's adjusted view. College counts sack yardage as rushing, as the NCAA does.
Other models what FPI, SP+ and Elo say about this game, beside Atlas and the market
| Model | Margin | Win | As of |
|---|---|---|---|
| FPI (ESPN) | PSU by 12.5 | 81% | Oct 28 |
| SP+ | PSU by 11.6 | — | Oct 29 |
| Elo | PSU by 18.3 | — | Oct 29 |
| Atlas | PSU by 11.1 | 76% | live |
| Market | PSU by 9.5 | — | live |
Each model's number for this game as its source published it before kickoff; Win is the favored side's chance to win, where the model gives one. FPI's margin and chance are ESPN's own. SP+ is the difference in the two teams' ratings, plus 2.5 for a home side; Elo is the difference in ratings at 24 points a point, plus 2.6 for a home side, as Atlas measured it on past seasons. None of these is an input to Atlas's number, and none is here to settle anything: they show where Atlas sits among other models.
How this grade was computed the four things that set the letter, out of a hundred
Solid card
A moderate disagreement in a band where the model's claim and its realized accuracy stay reasonably close.
Atlas sits 1.6 points from the market. The calibration curve, fitted to every completed season out of sample, expects cards at that distance to fall 6.0% short of what they claim — which is where the calibration component above comes from.
The season is mature enough for the team profiles to have settled, and the market has not moved much since it opened.
The grade describes how much weight the information deserves. It says nothing about which side of a market anyone should take, and Atlas does not publish that. How grades are computed.
All drivers all 4 things the model is reading, and how each team ranks against the rest of the country
Atlas rates Penn State's defense at +7.9 points a game against FBS average and Purdue's at -0.8, opponent-adjusted, after 7 games this season.
A combined 106 plays per game — the 35th percentile of this season's games. Few possessions, which pulls the projected total down.
Playing at home is worth 2.6 points in this season's model, fitted, not assumed.
Penn State allows a successful play 36.0% of the time (29th percentile — lower is better); Purdue 42.3%.
Market movement how far the market has moved since the number opened, and which way
Total, open to current
| Measure | Value |
|---|---|
| Opening number | 52.5 |
| Current number | 54.5 |
| Market direction | up 2.0 |
| Atlas direction | down |
Atlas and the market point opposite ways on this total. That is information, not a verdict.
Two findings that are not in conflict
Atlas has measured disagreement twice, against two different reference points, and they say different things.
| Measured against | What it predicts |
|---|---|
| The closing number | the model's own failure — so a large difference lowers the grade |
| The opening number | which way the line moves — tracked publicly, not graded |
A model far from the close is wrong; a model far from the open is early. The second finding is tracked in the open closing-line record and is deliberately kept out of the grade.
Reliability record what Atlas claimed and what it delivered, across the seasons it never saw while being built
Calibration by disagreement band
Two series, labeled directly. The widening gap is why Atlas grades large disagreements down.
This band, five seasons
| Measure | This band | All cards |
|---|---|---|
| Realized accuracy | 49.9% | 49.2% |
| Claimed accuracy | 53.8% | 58.0% |
| Calibration gap | −3.9 pts | −8.8 pts |
| Seasons above 50% | 3 of 5 | — |
| Games measured | 739 | 3,864 |