New set is out so time to look at very early data to get the first hints of what it's all about. Remember - we are talking about very small sample size and a relatively bad dataset as people still don't know what they are doing so treat with caution! 1/8 When it comes to color pair power level it seems we have 3 early tier one combinations: WU, WB and RB. There is a large chunk of color pairs that are lagging behind - remains to be seen which of those are just bad and which ones are just badly build in the early days. 2/8

Cube Season is upon us so a quick data thread. And it is about draft portion. Win rate data is still way too low in volume, but instead, I will focus on the pick rates of 17Lands users and general Arena population. 1/12 "But how do you get the Arena pick rate from non-17Lands users?" I estimate it. I know ALSA is non-linearly linked to pick rate and based on it, I can recalculate what the pick rate should be for given cards. Any outliers in the 17L user data can be then corrected for that. 2/12

Omniscience draft sounds like a coin-flip, but it isn't as it currently stands. If you think that who is on the play, wins - you are probably losing some value there. Also - you will probably have ~20% win rate against me. It took several drafts, but I think nailed it now. 1/9 First thing: the meta plan. Everyone figured out card draw is good. But this is only layer 1. Layer 2 is disruption. Counters and discard. Layer 3 is the inevitability. You want to build your deck so that after your turn oppo is locked. Hellbent against a counter or dead. 2/9

It is time to help constructed community with data and look at the RC DC metagame and win rates in Pioneer. Let's start with the tournament-wide win rates for the top 10 played decks. As you can see top 4 played decks did reasonably well, but MonoB was the real winner at 54%. 1/7 Notable flops were Greasefang and MonoG (not that @DanaFischerMTG cares - congrats). Also Rakdos Transmogrify and Jund Sac underperformed.

Of the top 4 played decks, Rakdos Aggro did best at 52.1% but Azorius Control, Phoenix and Enigmatic Incarnation were close behind. 2/7

Hot takes time! Everyone is loving Manifest Dread so time to put it down a notch. This thread I will air all my doubts about the new mechanic. Don't misunderstand me - I am not claiming it is rubbish but I think we shouldn't be forgetting downsides. 1/11 Tagged along many spells in the format, MD lets you look at the top 2 cards of your library and put one of them face down. If it is a creature, you can turn it face up any time for its casting cost. That is neat. But is it broken? 2/11

Today a bit of a deep dive into a never-explored draft data, so please give me comments on how you like it and what would be interested in this area. We go deep on game play stats dug out from the murky depths of replay data on 17L. And it is all about turn 5.

Why turn 5? 1/21 It is late enough in the game to have loads of data, but not so late that I have to exclude too many games. And I excluded every game that ended t5 or earlier, so keep in mind - all the most lopsided games are not in this analysis, something to keep in mind. 2/21

Early data can be misleading. And in MH3 we have a great case study for that.

Dimir.

On the face of it, it has a middling 54% win rate. In the middle of the pack of 6 lowest win rate archetypes. But scratch the surface and you might have a diamond in there.

1/6 Firstly, despite the decks winning only 54% of the games, individual cards do well. 8 commons have a GiH WR >59%.

Pairs with similar WR don’t reach this level. UG has 5 such commons. BR - 2. UR, BG and WB only 1. That is a significant difference.

2/6

WotC announced a new type of product that will combine Set Boosters and Draft Boosters. We don't have 100% info, but there is a significant chunk we do know. How will the new product impact draft? Not by much at first sight. 1/12 Firstly, WotC focused on the fact that you could potentially open up to 4 rares in a pack. Yes, theoretically it is true. But realistically - it probably is a once several boxes experience. To look at it, we need to see what would need to happen. 2/12

Main complaint about data in Magic are absolute opinions. "This card is busted because it has a high wr". You hear such arguments all the time. Using data as a hammer to quench any discussion annoys me too. This thread aims to look at win rate with a bit more subtlety. 1/19 First - let's rethink what a win rate is. In its simplified view it will frequently look like this: Titanic Growth has a good Game in Hand Win Rate in RG decks. It is the 4th best common in those decks so I guess it is just a good RG card. 2/19

March of the Machine is arguable a set with the most complex booster pack architecture. No dedicated rare slot, instead a double sided slot, single sided slot, Battle slot, land slot and a multiverse legend. How can this impact draft? 🧵 1/11 Thanks to the infinite kindness of @17Lands team I got some early access data to try and figure out what is the impact of the pack collation on the card frequency. Firstly let's look at how packs look like in a regular set: Simple 1 rare/mythic, 3 uncommons and 10 commons. 2/11

Often people claim to have a signature build, something others don't do in a Limited format. And often they are right. But such "special builds", sub-archetypes are notoriously hard to find using data. For many reasons. In Magic Numbers #69 I try to find such builds in ONE 🧵 1/x First difficulty is to find those builds. But over time I developed a toolbox to do it. And weirdly, I use methods from my day job to do so.

I am a microbial ecologist, studying how microbial communities function and which species make them tick. 2/x

If you wanted a worrying graph about ONE draft - here is one.

Most color pairs don't have commons that perform OK. I usually use a 56% GiH WR cut off to measure OK cards. In ONE 6 color pairs have <10 commons that have that WR. In BRO it was 0, DMU - 1, NEO - 2. 1/x Yes, part of it is because some archetypes have an appalling win rate. But that is a problem in itself. Bottom 4 color pairs in this dataset are all from ONE. It would be easy to blame it on players who just didn't figure out the right builds, and in part it is probably true. 2/x

Did you ever wonder what are the odds of opening that Vorinclex in ONE pack in Arena? Well...

It is ~57.5%. There - no clickbaiting. But if you want to know how to get that number out of @17Lands - explanation in the thread.

1/x @17Lands You need to look at the times a card has been seen in the Card performance data. Number times seen depends on number times open. But! Of course cards that go later will be seen even more. Just look at the difference between Staff of Compleation (2313) and Nissa (777) 2/x

Last Magic Numbers were about reading draft signals.



Here is a TL;DR thread of the seminar to communicate most important findings. Signals in draft are often mentioned in limited content but there is rarely any specifics. @17Lands can change that. 1/18 @17Lands Firstly: what is a signal. By signal I understand any information during the draft that let's me effectively judge openness of a color or color combinations. 2/18

The other catch'em all - Wretched Throng, performs surprisingly well in multiple archetypes, not only in UB. Worth keeping an eye out for them going late to speculate. And in this case - definitely don't play 1!!

8/x I also look at every archetype and identify best performing and worst performing cards for each. To see the details of methodology - watch the stream, but I basically look at cards overrepresented in well performing decks. 9/x

The VOD of yesterday's Magic Numbers is up. This time I look at general VOW numbers and initial look at what cards make each archetype tick.
twitch.tv/videos/1209322…

Some interesting finds in thread below:

1/x First of all, VOW is not a 10 archetype format - it is more like Ikoria, where enemy pairs were more supported and allied colours were not fully supported. Here it is vice versa, with allied pairs having better archetype support. Allied pairs are drafted more. 2/x

The VOD of last Magic Numbers is out!

twitch.tv/videos/1163774…

I talk about how to understand win rates and limitations of using this metric. This leads to the main issue with win rate imho: averaging card wr across a color combination hides cool, but niche sub-builds. 1/13 Imagine an archetype where there is the standard build, 90% of the people play and a niche, little known build-around, that 10% of people play. If a card has 50% wr in the standard build, and 70% in the build-around, data will show 52% total win rate. 2/13

I think it is time to talk about the Price of Loyalty problem. The card has absurd win rate. RB is good but a lot of its success comes on the back of this severely underdrafted common. And it is not likely to change as it only fits there as black has all the sac outlets 1/x Rakdos has 59% win rate despite being as much as 13% of all the @17lands user metagame - way above what it should be were the colours balanced. And Price has an even higher win rate - 62.5% when drawn. This level is reserved to top mythics. Prof. Onyx had the same GIH in STX. 2/x

The VOD of my seminar on the signal reading in AFR is out: twitch.tv/videos/1087818…

A little TL;DR to give you the impression on what it was about.

Reading signals in draft is important but can do some harm too if done incorrectly. 1/10 There are important caveats in reading signals:

- signal is based on community evaluation of a card rather than your evaluation;
- signals have strength and range;
- signals are ambiguous and there is plenty of noise in draft;
- signals can be based on presence/absence. 2/10

New D&D expansion #MTGDND brings d20 rolls toMagic. Here is how to look at these random effects and my thoughts on some of the cards. Firstly, majority of the cards have 1-9 as baseline effect, 10-19 as bonus and 20 as a large success leading to these percentages: 1/x Good Magic players tend to maximise deterministic outcomes, and even with die rolls, it is to some extent possible. To get there, Pixie Guide lets you roll more dice and chose the highest outcome: 2/x

A teaser from the big analysis of Lessons i am working on currently. I will focus on the colleges but there is one interesting thing I found in off-college archetypes: Pest Summoning as a secret start of the Dimir decks. 1/9 In the ~1200 Dimir deck games in the dataset I looked at, in
roughly 50% players had access to Pest Summoning in their SB. And that made all the difference. Decks without it had a paltry 48.7% win rate, decks with - 57.1%. 2/9