My computer says the expected point total of the Virginia/Purdue game is 98 points. That’s the first time the computer has forecasted a total below 100 in the 30-second shot clock era. But it’s almost certainly too low.

As of this morning, sports books are posting the total at 103. As far as I can tell, this is the lowest total from the past seven seasons. This beats a week-old record from UVa’s November 27th game against Maine where the total was around 105.

So if one’s model is forecasting a total below 100, one’s model has a blind spot. Of course, college basketball games can and do fail to make it to the 100 mark. Virginia has already played in four of them. But your model just shouldn’t ever expect it.

And if you can’t expect it in this game, then you can’t ever expect it. Virginia has a great defense and a bad (for the ACC) offense. Purdue’s identity is less obvious, but at their best the Boilermakers have also played great defense, and their offense has some issues. Furthermore, while Virginia once again has a stranglehold on the slowest pace in the land, Matt Painter’s team is inexplicably ranked 348th in adjusted tempo (a.k.a 6th slowest). Painter’s had just three teams rank outside the top 200 in this category in his life so this is unusual for him. This team just doesn’t have any way to score quickly.

Throw in the fact that offenses in general have struggled this season and you have the perfect storm for a low-scoring bloodbath that, with a more modern set of rules, would not be possible in college basketball. Nonetheless, the Vegas total is a little bit higher than my model, and if one is interested in predicting things, one should wonder why one’s model is missing something.

I’ll offer some guesses for this case.

First, we know that when two slow teams get together the resulting tempo is usually slower than each team’s average. But in extreme cases, this relationship starts to break down. Due to the constraints of the shot clock, the game can only get so slow. Virginia’s adjusted tempo is 58.4 but put another way they like to play 11.1 possessions slower than their opponents. But Purdue plays at 62.5 which means this game should be played at 51.4. But that’s not a realistic expectation.

My method for predicting possessions already attempts to account for this issue but it might not be doing enough in this case. The Virginia-Purdue game is forecasted to have 54 possessions but neither team has played a game that slow so far. Purdue hasn’t had a game with fewer than even 63 possessions, though in fairness they’ve played a heavy dose of fast-paced opponents.

A 54-possession game requires an average of 22.2 seconds per possession. That’s not inconceivable, but it takes a special blend of offensive resistance to transition and a lack of silly turnovers immediately after rebounds that can inflate the possession count. Oh, and a bunch of offensive rebounds would be nice, too. It also helps to have a close game as any sort of significant lead will result in one team taking shots a little earlier in the clock and committing fouls late.

In fact, there hasn’t been a game this season that slow. And there were just six games played that slow between the beginning of the season and mid-December over the previous four seasons. So we might guess that this game will be a little faster than the computer thinks.

Another possibility is that bad offense can only get so bad before it helps the other team’s offense. It’s easier to score after an opponent’s missed shot than when you have to take the ball out of the basket. While this effect exists it’s probably pretty small. It hasn’t seemed to matter for Virginia so far, especially.

The third thing I can think of, which mainly applies to Virginia is that any outlier is bound to regress to the mean a little. It’s part of the reason the weight of the pre-season estimates needs to be strong to get accurate predictions this time of year. With respect to Virginia, their defense isn’t as impenetrable as the early-season numbers indicate and their offense won’t truly suck as much going forward as its early performance would lead you to believe.

Teams with a large split in the offensive and defensive numbers tend to see those numbers converge as the season progresses. Virginia’s offense hasn’t had to be good so far and it probably never will be. But eventually the defense will face teams making plays and making shots and the Cavaliers’ offense will inevitably work a little harder to respond. It’s likely not a coincidence that UVa’s best offensive game in terms of points per possession came in their worst defensive game (a 61-55 win over Vermont).

And a look through history also confirms that ever expecting a game to go below 100 points is a model bug. Those that think this season’s Virginia team is unique clearly did not live through the 2005 Washington State season. The ’05 Cougars were first in defense and 264th in offense to go along with the fourth-slowest pace. Even with that unwatchable combination and a longer shot clock, just three of Washington State’s 28 games ended up with fewer than 100 points scored.

The last time my computer predicted a sub-100 game was on November 19, 2014 when it had Columbia beating American 48-44. (Columbia won 52-43.) But if the computer was doing a better job, it wouldn’t happen again. If you forecast a game under 100 points, you are going to be on the low side more often than not.