Yes, took me a while to understand my misconception. I went back to the data and adjusted for that and the differences were gone.
Sorry for the confusion and thanks for bringing the issue to my attention!

Just to be clear, the "unexpected" thing I found turned out to be a mistake in my analysis. The wrong parts of my analysis were removed from the article to avoid spread of misinformation.

I'm truly sorry for the confusion I was creating, and I'm proud of our vigilant community to spot such a mistake early on.
I removed the sections about the rank distribution and pod seatings from my blog post, the rest of my article about the 17Lands draft share is still valid though.

looking into them I found a fundamental mistake I made in my analysis.
I assumed that the distribution of ranks within the 17Lands drafts was constant but of course it changes drastically each month. Looking at smaller time windows of only a couple of days revealed no significant differences.

I definitely want to do this going forward. Not sure if I'm gonna do it for past sets, it took quite some time to load all the data into a database and run the matching algorithm.
Maybe if I find some more performance optimizations I will consider it. It's for sure interesting!

The rank data is captured at the time of the draft. So the pairs of paired ranks is always a current snapshot. If I understand correctly what you are saying this should not be an issue then, right?

I did a trophy deck analysis of Orzhov because I also felt it was underdrafted. I think there’s potential for some fun decks.

state-of-the-format.com/posts/orzhov...

Real nice statistical take on the self-correcting nature of draft. Thanks for crunching the numbers 🙏