Rk9

Telling time is such a banger, name, effect, flavour text are all 10/10.

🧵 Announcing Ursa's Research Desk - a flexible dashboard for analysing MtG tournament results and deck performance brought to you by team @busstop-mtg.bsky.social and MtG Bazaar!
https://arckaynine.github.io/Urzas-Research-Desk/

9/ The project also would not have been possible without the support of my friends and team mates (@busstop-mtg.bsky.social) and our sponsor, MtG Bazaar. I hope you enjoy your own exploration of the data, let me know if you find anything interesting!

8/ particularly the analysis done by @jkyu06.bsky.social and @ajlvi.bsky.social, the data work done by @sierkovitz.bsky.social and mtgds.bsky.social, and of course the inimitable @frankkarsten.bsky.social and @fireshoes.bsky.social

7/ This project has been built on the fantastic work of many people, including fbettega's work to continue the data collection efforts of Badaro (https://github.com/fbettega/MTG_decklistcache), and has been inspired by the work of many others, ...

6/ Urza's Research Desk is built with #python, #panel the #holoviz ecosystem. It is deployed with pyodide and GitHub pages. You can see all of the code in the tcg-research-desk repository - and try it yourself by installing the library from pypi!
https://github.com/ArcKayNine/tcg-research-desk

5/ We can also see the matchup win rates for each archetype in the "Matchup Matrix" tab (hovering over each value will provide the 95% confidence interval for a matchup):

4/ An interesting split between builds of the archetype is that some PT teams cut metallic rebuke. With the "Card Performance Analysis" tab we interrogate how they did and see some indications that trimming them might be the way forward (error bars are 95% confidence intervals):

3/ Ursa's Research Desk can help you answer these questions and more! Start by selecting an Archetype to see what people are playing across all MTGO and Melee data - for example Affinity, which had a variety of builds show up at the Pro Tour last weekend:

2/ This project grew from wanting to answer some of the specific questions my team had leading up to the Pioneer RC last year - specifically how many Artist's Talent to play in Phoenix. After writing some python I wanted to provide a way for my team to explore the data themselves.

🧵 Announcing Ursa's Research Desk - a flexible dashboard for analysing MtG tournament results and deck performance brought to you by team @busstop-mtg.bsky.social and MtG Bazaar!
https://arckaynine.github.io/Urzas-Research-Desk/

What a sicko

I almost certainly got my current job as a direct result of being able to sell the former as an asset of mine.

Reposting for other time zones. Might be interesting for folks like @lordoftheboard.bsky.social, @spacecatspod.bsky.social, and I know the @loadingreadyrun.com crew enjoy a game of root every now and then.

Back to the Woodland: A Retrospective on the Marauders Era of Root Welcome back fellow woodland warriors! It's been a while since my initial blog post trying to understand the complex faction interactions in...

I continue to be fascinated by @leder.games’s Root, and have written up an analysis of 9,000 games played by the Digital League community:

compulsiveresearchmtg.blogspot.com/2025/08/back...

YouTube video by Team Bus Stop Australian 7 Point Highlander | Lurrus Breach | Deep Dive | Team Bus Stop

🚨Even more Australian 7pt Highlander content alert 🚨

The team's resident combo enjoyer @brennancrawford.bsky.social sits down with
Steven Hitchcock to talk about one of 7ph's premier combo decks: Lurrus Breach!

www.youtube.com/watch?v=VCYx...

MainDeck 5 Island 4 Opt 4 Stormchaser's Talent 4 Burst Lightning 4 Monstrous Rage 3 Into the Flood Maw 3 Mountain 1 Spell Pierce 4 Cori-Steel Cutter 4 Stock Up 4 Riverpyre Verge 4 Spirebluff Canal 4 Shivan Reef 4 Sleight of Hand 4 Monastery Swiftspear 4 Slickshot Show-Off Sideboard 1 Disdainful Stroke 2 Abrade 2 Torch the Tower 2 Ghost Vacuum 1 Witchstalker Frenzy 1 Into the Flood Maw 2 Drake Hatcher 2 Ral, Crackling Wit 1 Spell Pierce 1 Lithomantic Barrage

Lost my win and in for top 8 and the PT at the ANZ RC today, finishing 17th. Quite happy with how things played out, slickshot main was the right play. Congrats to @bentudman.bsky.social for making top 8, keen to watch him repping @busstop-mtg.bsky.social tomorrow!

Had a great time yesterday running the conclusion to my Ravnica DnD campaign, involving this over-engineered shadow puzzle that required a scavenger hunt across my bookshelves (mirroring in-universe bookshelves). Very proud of how it came together.

My friend with his Thassa’s Oracle/Doomsday combo in sealed.

Despite playing magic all weekend and losing a bunch of modern matches, it took until the mystery booster sealed for me to actually lose to Thassa’s Oracle.

I believe in you! Could always go for a third screen if you’re getting impatient.

Pretty easy when you can’t see without glasses, just stare at the blur of your reflection and wait it out.

I had wandered if GW being on top was because of outliers.

Ah, yep, that makes more sense. The takeaway then is not consistency on a set by set scale but that modern limited design sensibilities have only a slight bias towards certain pairs, but are consistent overall.

It’s wild to me how temporally consistent these are from set to set. I’d expect a lot more shifting and outliers across the archetypes.

Emrakul, the Promised End doesn’t count?

Well past time for my Annual*** Ravnica Rotisserie draft (RAV, GPT, DIS, RTR, GTC, DGM, GRN, RNA, now with MKM and RVR), but better late than never!
We're off to an interesting start, with MKM and RVR additions shaking up the metagame. I'm in Seat 4 with an OK start for a Selesnya aggro/mid deck.

MNIST embeddings trained with the 2 class removed in UMAP and Parametric UMAP, with new data transformed in that includes the 2 class. The updating parametric UMAP successfully creates a new cluster, in comparison to the other methods which blur the 2s in amongst the other points.

A recent new addition to umap-learn was the ability to have ParametricUMAP embeddings that can be smoothly updated with new incoming data -- creating new clusters as needed, and gently adjusting the rest of the embedding to make space for them.

umap-learn.readthedocs.io/en/latest/tr...