Prediction
We Went 4-2 in Week Zero. The Pick We Liked Most Is the One We Blew.
Three days ago we published picks for all six Week Zero games and said we’d show you how they did. Here they are.
4-2 straight up. 3-3 against the spread.
The scorecard
| Game | We said | Actually |
|---|---|---|
| North Carolina at TCU | TCU 28-17 | North Carolina 15-10 |
| San José State at USC | USC 42-10 | USC 42-26 |
| NC State at Virginia | Virginia 24-23 | Virginia 34-8 |
| Hawai’i at Stanford | Stanford 24-23 | Stanford 37-27 |
| New Mexico State at Florida State | Florida State 38-14 | Florida State 34-17 |
| Memphis at UNLV | UNLV 28-27 | Memphis 27-21 |
Two numbers on the model itself. It called 66.7% of winners against a backtested 70.4% — in line. And the scores missed by 7.67 points per team against a backtested average of 9.09, which is better than it had any right to be.
The one we got wrong, and why
We wrote this about North Carolina–TCU: “The one game where we’re more confident than the market.”
North Carolina won 15-10.
That was the single pick where we told you we saw something Vegas didn’t. It was the most confident non-blowout call on the board at 75.3%. It was wrong, and “the model liked TCU” is not an explanation.
Here is the actual one. The model’s own factor breakdown credits TCU 2.8 points for the distance North Carolina had to travel.
That game was played in Dublin. Both teams flew to Ireland — and Fort Worth is farther from Dublin than Chapel Hill is.
The model correctly stripped out home field for the neutral site. It docked exactly 2.8 points for that. Then it handed the same 2.8 straight back as a travel edge that did not exist, because it measured the trip as though North Carolina were visiting Texas.
Net effect on a game decided by five points: we spotted TCU nearly three points it never had.
That is a fixable input, not a bad read. It gets fixed before Week One.
The one we got right when it counted
We also told you we disagreed with the market on San José State +38.5 — that USC would win comfortably but not by that much.
USC won by 16.
That is the pick we’d point to. Not because the winner was ever in doubt, but because the number was, and we said so in public three days early.
The rest of it
Both strong picks landed — USC and Florida State, and Florida State was our most accurate score of the week at 3.5 points per team off.
The three toss-ups went 2-1, which is exactly what a toss-up should do.
The two we got wrong were the two closest to a coin flip in the model’s own head — and one of those, Memphis at UNLV, still graded as an against-the- spread win. UNLV missed a field goal and gave up the last 13 points.
The part we’re not going to spin
Our solid tier went 0-1. Our toss-up tier went 2-1. The confidence labels ran backwards, and the model’s own tier check flags it.
Six games is nothing. One game in a tier is nothing. That break is almost certainly noise and we’d be lying to tell you otherwise — but we said we’d publish the grade whether or not it flattered us, so there it is.
The honest read on Week Zero: the model called two blowouts correctly, called three coin flips about as well as coin flips can be called, and lost the one game it claimed an edge on for a reason that turned out to be geography.
Next
Week One, and Nebraska is finally in it — Ohio, Saturday September 5, 11:00 a.m. in Lincoln on FS1.
Husker Nation, we’ll have the full slate picked before kickoff. And the travel math will know where Ireland is.