Seek New Knowledge Art by Dave Greco

First Look at Alchemy Horizons: Baldur’s Gate Draft Data: The Big Picture

There are different ways to evaluate a limited format early. You can put in the reps and use your own reasoning to assess the card power. You can watch streams of others, lean on their judgement. Or, like I like to do, look into 17Lands data and try to figure out from those numbers how does the format shape up, what makes each color pair tick, and which cards are traps.

In this article you can follow my train of thought to learn about Alchemy Horizons: Baldurs Gate (HBG) draft but also on how to interpret the data available.

Dealing with Data

Data is a useful tool to help you become a better Magic player but it comes with the same reservations as any other method of getting better. And rule number one is: “Be careful”. If you evaluate cards badly, or fall into a trap of watching a streamer who has no idea what they are doing to hone your skills, the result may be opposite than what you intended to achieve.

Why would that be any different with data? If you look at the data without taking context into account, without multiple caveats and without remembering about many caveats and careful interpretation, you might arrive at unfounded conclusions and harm your skill rather than improve it.

This applies especially to the early data, when sample size, number of observations related to each card is still low, which means that natural variance related to luck, randomness of Magic, can still play a role. This is why before we dive in, I ask you not to treat those numbers as absolutes, but as very useful guidelines but with a margin of error.

There is another important caveat for early data. At the beginning of the format we are still trying stuff out and, necessarily, the data contains those trials. Trying new things can muddle the picture as successes will be interwoven with failures, so some successful strategies will only become apparent later in the format.

I find it best to look at the broad picture first. What removal spells are available? How does the format look like? How fast is it? Which color pairs win the most? Knowing answers to those questions helps you frame the results of individual cards in the context of the format better. It makes it easier to understand why did a card you thought was good underperform compared to expectation and why a card you thought was poor become one of the format’s all-stars.

Two Color Format

Data suggests that HBG is a two color format. Not only >90% of the decks from 17Lands users are predominantly two colored (~22% of them with a minor splash in 3rd color), but also there seems to be a win rate penalty for including multiple colors. And this should not come as a surprise. There is no land color fixing, which usually is essential to play more than two colors. Instead we see either treasures or creatures that give mana of any color, useful tools for splashing but notoriously tricky to apply to three color decks.

Looking at the win rates for each color numbers in deck, you can see a decrease with addition of extra colors. Mono-coloured decks win the most, but this is a very atypical situation to happen. Only rarely will you be in a seat that allows drafting a viable mono-coloured deck. And unless that happens, you should be trying to draft a two-color deck to maximise your win chances. I know splashing can be tempting, but in most cases it is better to miss some deck functionality (like having removal) and stick to two colors and be able to reliably cast all your spells, than splash it, gain a small edge through that but lose consistency.

17Lands users preferences of number of colors in a deck (A) and win rate based on the numbers of colors in a deck (B)

17Lands Users Love Mardu Colors

OK, you know now that you should play two colors. But which ones? A good way to start is to check what do other people play and, importantly, are they successful playing it? In a perfectly balanced format, each color pair should be played exactly 10% (as there are 10 color pairs) of the time you play a two color deck. One look at the chart below and you can see 17Lands users don’t draft as if HBG was a balanced format. Quite the contrary.

55% of all their games with decks of 2 colors are with WR, BR and WB decks, so all the two-color combinations from the WBR Mardu wedge. This is almost twice as much as you would expect in a balanced format. On the other side of that spectrum, only 18% of the two color decks had blue in them. That is less than a half of what you would expect from a balanced format.

Now this is based on only 17Lands users, which are only a part of the whole metagame. This does not necessarily mean that the draft metagame would be so imbalanced looking at all Arena users. But 17Lands users also have a higher win rate than average (56.5%), which means their preference brings in the results, which is probably a good reason not to ignore them. Apart from Mardu colors, only WG color pair is played more frequently than the default 10%. Rest stay well below, which means that if there are powerful builds linked to them, you can potentially find an open lane towards them. But that is a big if.

Why you might ask? The most drafted colors are also the ones that win the most. Top four color pairs in terms of win rate are… You guessed it – three combinations of Mardu colors and WG. Some way behind them in 5th there is the 5th most popular color combination in BG. The other combinations win significantly less, with Izzet (UR) miles behind anything else.

Does it mean you should force drafting WR and WB? Not exactly. I mentioned that early format data is a bit odd. One particular trend I observed across multiple formats so far is that the early metagame is shaped by which deck is the most intuitive to build. In most cases those are the ones where what WotC suggests we should be doing aligns with what is successful or decks based on sheer card power. It takes a much longer time to find the best builds for synergy driven non-obvious decks, but many times they become very successful in the later stages of the format.

17Lands users frequency of playing each color pair when they play two color decks (A) and win rate of each color pair (B)

Speed of the Format

It is important to know just how fast any format is. Having this information lets you prioritise cheap, for example, creatures and interaction (fast format) or card advantage engines and mana sinks (slower formats). In fast formats, as a rule of thumb, it is much more important to be on the play. The shorter the games, the bigger the advantage of having the early initiative. But as the games get longer in a format, the advantage of the extra card you draw while being on the play becomes more important.

HBG is, according to the 17Lands data, the fastest draft format we had so far on Arena. This also leads to it having the biggest on-play advantage in terms of win rate. Games in HBG Premier Draft (BO1) last on average 8.8 turns, and if you played with an opponent of equal skill, you would win 53% of the games you are on the play. Both those numbers are unprecedented for a standalone set in Arena’s history.

On average, games in limited formats, looking at sets since Ixalan, last ~9.4 turns. In some slower formats, like Theros Beyond Death or Ravnica Allegiance, even longer, approaching 10 turns. The difference between 8.8 and 10 does not seem large, but yet somehow we intuitively understand which format is considered fast and slow without a miss without knowing those numbers, so this “small difference” is perceivable by players in how the format plays out.

On play win rate advantage is another metric where HBG is a historic outlier. 6 percentage point (%p) difference between being on the draw and on the play is a lot. Typically we see the difference between being on the play and draw at 3%p level, meaning you win ~51.5% of the games on play and 48.5% of the games on play. And slow formats can even promote being on the draw. Only two formats in recent memory that had a higher win rate on the draw were very slow Ravnica Allegiance and Theros: Beyond Death.

HBG being so fast puts very real constraints on deck building. You can’t assume you will survive till late game if you don’t focus on early game interaction. And even if your deck is well prepared for the early onslaught of aggro strategies, you can still lose to a very good curve-out. At the same time, even if your deck is pretty aggressive, you can’t assume you are the beatdown. You should always assess what is your status in the game and play accordingly. Attacking optimistically when it is your opponent who plays the more aggressive aggro deck can be a costly mistake.



Average game length and “on play” advantage in recent draft formats based on 17Lands data

Openness of Archetypes

In draft you should actively look for open color pairs and try to draft them, because doing so maximizes your chances of drafting solid cards. You can do it by being very receptive during the draft and reading signals, but that is not all you can do. You can come partially prepared. Some color pairs in the format will be on average more open, others less. This is based on two main factors. Firstly, Arena users have drafting preferences. You already saw that 17Lands users draft Mardu color pairs more often. This, but taken to the level of all Arena users can help you estimate how aggressively is a given color drafted.

Secondly, we can look at the color pairs power level. Colors with more powerful cards will be naturally deeper and can serve more players in the pod. This also means that pretty likely it will be more open on average than a color that can only serve one drafter per pod, at least in the early stages of the format.

To see which pairs are more open than others I devised a simple method of estimating the average color pair openness. I look at how many cards with Game in Hand (GiH) Win Rate over 56% are there opened on average per pod for each color pair. GiH Win Rate is one of the most important metrics on 17Lands. It gives us the most information about the power of an individual card, by looking only at the games where it was drawn and ignoring those when it was not drawn. This way it lets us measure the win rate only in the games where the card should have some sort of impact – either by being good and helping you win or quite the opposite, by being detrimental/not castable.

56% is a decent win rate, so looking at cards with higher win rate I ignore cards that perform badly. This also means that there are many cards that fulfil this criterion for color pairs with high win rate, and few for color pairs that win little. That way my measure of openness takes into account the power of each color pair on top of availability of individual cards. It will happen often that a white card, which has a GiH win rate > 56% in Boros (WR), will be under the 56% threshold in Azorius (WU).

To calculate the openness I don’t only need the list of cards that fulfil the criterion but also to know how many I will see per pod. To do it, I look at ALSA, Average Last Seen As metric from 17Lands, which tells me how late on average is a given card picked. ALSA of 1 means you only see this card as a first pick, you literally never get it passed. As the number increases, it means you get it passed more and more. If ALSA is >8, a card wheels regularly.

Since we know how many cards of each rarity we open per pod (for example ~2.4 of each common are opened per pod), from ALSA I can estimate how many copies of each card will be seen by 17Lands user if sat at a table with average Arena users. If I add up those numbers for each card with GiH win rate >56% for each archetype, I get the average number of good cards you see in a draft – the color pair openness.

Openness of each color pair in HBG

If you look at the openness data, you see that three color pairs with highest win rates in the format are the most open. For example, you will see 53 Orzhov cards with GiH win rate >56% in an average draft. This means that in multiple pack you will have a choice of several good cards, giving you flexibility in your draft. Same goes for Boros and Selesnya. Having a choice will let you draft a better plan for the deck, than being at the mercy of picking the only playable in the pack.

Next three colors, Golgari, Rakdos and Simic are not far behind (37-42 cards per pod), but only one of those combinations has a high win rate, Rakdos, the other two have more medium win rates. When I look at the data and see that a color is fairly open, but yet that openness doesn’t translate into a high win rate, one potential explanation is we build the decks of that color pair sub-optimally. In my opinion, based on this data, there is a lot of space to innovate in Simic and Golgari decks and as the format matures and top archetypes will become more contested it is those two pairs we should look for improvements.

After the top 6 color pairs there is a significant drop-off in openness. WU still has some good cards opened per pod, but fewer than needed for a functioning deck. This means that even if you find an open WU lane early, by chance you may not get enough playables if wrong cards were opened in the pod. It gets even worse in UB and RG, but even their bad results bleak in comparison with Izzet. UR will have only 0.7 good cards per pod on average. The reason? There are not many good card in this color pair. barring a handful of rares and uncommons. This makes my default position on Izzet to avoid drafting it.

However, openness is not a measure you can measure once and draw conclusion from it. Openness of a color pair is a dynamic state. If Arena users start drafting Boros cards higher, openness of Boros will drop. If we find and popularise more coherent builds of Simic decks and more cards will reach the 56% GiH win rate threshold, Simic openness will increase. I will be monitoring the changes in this metric to keep an eye out on the Limited metagame in the coming weeks. And predict the differences between the top six color pairs will flatten somewhat.

Conclusion

Next week we will take a peek in individual card performances – which cards are stars of archetypes, which ones are traps you should avoid. But until then, if you are interested in the topic, a more detailed version of this article is available as a video seminar below:

All data used here comes from 17Lands, for whose help and trust I am extremely grateful.

Card Kingdom - Double Masters 2022
Default image
Sierkovitz

I am a limited player, who mainly skips playing in order to analyse the limited data using 17Lands.com. I run a podcast: Magic Numbers, where I try to use data to let you improve your limited game play, find out which heuristics work out and which common ideas are not well supported by data.

Articles: 4