McCarville’s crazy comeback


In the Canadian Olympic pre-trials last weekend, Krista McCarville trailed Mackenzie Zacharias 7-3 heading to the tenth end. The game wasn’t televised but the tenth end looked like this in text form:

Even with lead Sarah Potts apparently hogging her second shot, Team McCarville scored four – and then stole one in the extra end to secure a very unlikely victory.

How unlikely? Well, I consulted my true win probability grid for the answer, and it said there was zero percent chance of winning when down four in the final end. Like, no decimal points or anything. Obviously, true win probability is not telling the truth, but it wasn’t that far off, either.

In serious women’s curling events over the last five seasons (events worthy of being included in my ratings and excluding junior events) there were 731 cases of a team trailing by four with hammer heading into a final end. The only known case of a team winning in that situation was in the 2020 USA Women’s Challenge Round when Christine McMakin beat Ariel Traxler by scoring four in the tenth and the nearly-as-rare steal of three in the extra for a 12-9 win.

So 1 out of 731 is zero with some small numbers after the decimal point. But it’s a good assumption that a comeback like this should be slightly more likely in the five-rock era, which didn’t begin until the 2019 season. However, it’s also true that this sample includes situations mostly against teams not as good as Team Zacharias, which ranked 22nd in the world heading into this event. It’s safe to say the literal “true” win probability in that spot is much closer to zero than one percent.

To further illustrate that, there were just two other cases in those 731 games of a team even forcing the extra end. And surely there were cases where the opponent simply conceded because the situation is practically hopeless.

It would be nice to know how often a concession happened but unfortunately that’s impossible. What I do know is that there were just 36 cases where a score was recorded in the final end. That doesn’t help much because not recording a score could mean either there was a concession before the end started or the final end was played until the losing team was run out of rocks.

So all we know is that 36 times the final end was played out to the final rock. And that’s an undercount to some degree as in most sub-national events there is an occasional lack of diligence in recording the score of the last end. Anyway, I’m going to be interested in seeing how often teams concede in this situation going forward.

There’s another piece of evidence to illustrate how improbable this was. During the pre-trials, Team Zacharias threw 48 shots labeled as “clearing” that weren’t tick shot attempts. Just four of them were scored as zeros and all of those were in the tenth end of this game. (In fairness, I strongly suspect the two shots labelled “clearing” for the third and fifth shots of this end were throw-throughs and simply mislabelled by our volunteer scorer.)

Improbably, McCarville punched her trials ticket in the following game by overcoming a 5-1 deficit after two ends to Jacqueline Harrison.

That comeback wasn’t quite as crazy, as true win probability says that a team has a 5.8% chance of winning in that spot in a game between two top 25 teams. (Harrison was ranked #45 coming into the event, so McCarville’s true chances were a bit higher.)

In other news, simulations for this week’s GSOC Boost National are up on the forecasts page. Somehow, Team Jacobs is still ranked #1 in the “official” world team rankings despite King Bruce Mouat finishing runner-up in the worlds and winning all three slams in the last seven months. Anyway, the King is the rightful favorite in my simulations, while Team Tirinzoni is the favorite in a completely wide-open women’s bracket.

In fake gambling news, I lost $2.1M (where M means either millions or made up sums of money) on my three bets on the GSOC Masters. This week, the computer says to continue to back Homan (15.6% chance to win, +750 at and Gushue (19.9%, +565). So we’re putting $1M on each.

Also, I pushed over version 1.2 of the ratings algorithm. Previous weeks have been backfilled with this update. It handles teams with limited results better than v. 1.1. I don’t anticipate any further updates for the remainder of the season.