The Data You Need to Crush Brother’s War Limited

Brother’s War has been around long enough for some real insight, from Magic Data Science.

The Brothers’ War has been available on Arena for just about two week.s What can we learn about the format from the data accumulated so far? I think the answer is: quite a bit.

That’s because 17Lands, as of this writing, has tracked 293,251 games. That means that frequently-maindecked commons, like Excavation Explosion, Scrapwork Mutt, and Evolving Wilds have been run in over 100,000 games, so the standard error of their win rates are on the order of 0.1%. This doesn’t mean that interpretations of those win rates aren’t potentially prone to certain biases (well documented here, and elsewhere), but it does mean that the literal facts aren’t much in dispute.

In fact, I was curious about how early in the format we can start to rely on win rates, so I looked at the public datasets released by 17Lands for each of nine recent formats. Those data sets have been released anywhere between 21 and 70ish days into the duration of their respective formats, but I’ll assume that the win rates at the time of release are Ground Truth.

If you take Game-in-Hand win rates calculated on day t of the format, and calculate the correlation of those win rates with the “end-of-format Ground Truth”, you get patterns like those shown below.

Basically, after about a week of data accumulation, correlation with “true” win rates is above about 0.75; after two weeks, it’s typically been over 0.9.

Join CFB Pro now!

Already a TCGplayer Subscriber?
Login with your TCGplayer Account to read this article now!


Leave a Reply

Scroll to Top