17Lands: In Defense of the Data
Don’t Blindly Trust the Data, But Do Believe It
This article is mainly a response to Chris Kvartek’s article here, and I recommend reading it before reading this article. If I had to sum up Kvartek’s article in my own words, it would be “Don’t blindly trust the data”. I agree with that sentiment, but I feel Kvartek undersells the value and the quality of the data you can get from 17Lands. I want to stress: the better drafter you are, the more you should be willing to trust your own judgement. However, even the best drafters should still trust the data, and try to learn from it.
In his article, Kvartek touches on how the data is made up from a range of player skills, implying that it makes the data less trustworthy. While it’s true that the data comes from a wide range of skills, I think that breadth is the data’s biggest strength – especially if you’re drafting on Arena. While draft games get matched by rank, the draft pods are the first 8 players to queue up, meaning you get a mix of player strengths. And, even if the breadth of the data isn’t convincing, 17Lands drafters are often ahead of the curve. The average win rate on 17Lands is 56%, which is a pretty big step up from 50%. In terms of games, 50% is going 3-3 in a draft on average, while 56% is just shy of going 4-3.
The 17Lands Stats
Before I go on, I want to quickly cover what some of the terms on 17Lands are, what they mean, how I use them, and how they might be biased towards certain playstyles/cards. Hopefully this helps you get a little bit more out of the data than just hoping that “bigger number is better”.
- GIH WR – Game in Hand Win Rate
- This stat is the win rate for the card for all the games it’s in your hand. It can be in your opener, or you can draw it later and importantly it doesn’t have to get cast to count.
- This is my preferred stat for evaluating a card, as I think it’s the least biased of all the stats, but it still does bias towards cards which are in more controlling decks, as longer games get to see more cards.
- OH WR – Opening Hand Win Rate
- This stat is the win rate for a card if it appears in your opening hand. If you draw it later, even on your first turn on the draw, it doesn’t count towards this.
- I use this stat from time to time, generally compared with GIH WR to see how a card “falls off” in a longer game. Naturally, this stat biases towards cards put in aggressive decks, and cheaper cards.
- GP WR – Game Played Win Rate
- This stat is the average win rate of decks that played a card maindeck.
- This stat tells less about the card, but more gives an idea of the strength of the decks it’s played in.
- GND WR – Game not Drawn Win Rate
- This stat is a card’s win rate if it’s in your deck, but you don’t draw it during the game.
- This stat is effectively the inverse of GIH WR, and I don’t often use it.
- IWD – Improvement when Drawn
- This stat is how much your win rate increases when a card is drawn, which is the difference between GIH WR and GND WR. If it’s negative, that means drawing it hurts your win rate.
- This stat can be pretty volatile, so I only consider it if it’s pretty extreme.
- ALSA – Average Last Seen At
- This stat is the average of the last times people saw a particular card. In other words, you likely won’t see the card after the pick number of ALSA.
- This stat is one of the least biased, since it has to do with how other people pick cards, and not how other people play. If you see a card well after ALSA, that may be a sign that lane is open.
- ATA – Average Taken At
- This stat is the average pick where a 17Lands user takes the card.
- Mainly, don’t confuse this with ALSA, but if a card is being taken a lot higher or lower than where you take it, it may mean you’re under/overvaluing that card.
One of the stats I pay most attention to is ALSA, since that can help inform when a lane is open, or if a card might wheel. 17Lands provides this tool which shows how the ALSA changes overtime, or, in other words, how the Arena population is changing its pick order over time. I find looking at the trends in this tool gives a better idea about card strength than any other stat alone. Just check out how quickly people warmed up to Organ Hoarder and Skaab Wrangler, while other cards hover where they are, or even start to dip as people pick them less.
One final thing about reading the data is making sure you’re choosing a good benchmark to compare with. As I mentioned, the average win rate on 17Lands is 56% and floats around that number, so a colour that has a 50% win rate is actually below average. Similarly, if you filter cards by how they perform in UW decks specifically, you want to keep the average win rate of UW in mind instead of the average win rate of all decks. Basically, don’t assume your baseline is 50%, and make sure you pick a good one.
When it comes to the colour rankings, I agree with Kvartek that the data for the mono-colour decks isn’t very useful. If you’re doing some wild stip-draft with your friends, then you might want to know which colour is your best shot, but being in one colour isn’t something to expect in a draft. In most sets, I find that only the data for the well-supported archetypes is meaningful.
With that said, let’s talk about Izzet in MID. At time of writing it has a 53.2% win rate, and is still 9th out of the 10 colour pairs. Admittedly, I was surprised to see it so low when I read the article, but after some thought, it makes sense. I do think that a well built Izzet deck is better than a well built Golgari (7th out of 10) deck, so why is Izzet lower? Like above, my answer is deck complexity. To make a good Izzet deck in MID, you need to be careful about your spells vs. your creatures, your interaction vs. your threats, and you really need to get there on spell payoffs. Golgari’s plan is simpler, but much less synergistic – play creatures and back it up with black’s plethora of removal. Because of the cards you need to balance, if you trainwreck your Izzet draft you risk having a deck that doesn’t function at all. If you trainwreck a Golgari draft, you have a bunch of creatures and removal – not ideal in this format, but still playable.
So, when I use this data, I don’t use it to say “X is better than Y, and the data shows it”- it’s not helpful, and it isn’t even necessarily true. If a deck has a lower win rate than another deck, or one lower than I’d expect, I take a step back and consider why. Is it a harder deck to draft? Does it rely more heavily on its uncommons, so the right cards have to be opened? Are the colours more contested, or not as deep? Is the deck harder to pilot well? Are people building the deck wrong? All of these questions bleed into the data. If you have answers to those questions, then maybe you can make some better decisions in your draft. But without the answers to those questions, I still think the data is meaningful. If a deck has an appreciably larger win rate than the rest – even if you may not know why – you should trust that there’s something good happening there.
To wrap this section up, I don’t use the win rates like a tier list. I use it more to measure my expectation of how a colour pair might perform; a map which gently nudges me towards certain archetypes and away from others. I want to emphasise nudge here: Don’t go and force Azorius every draft because it has the highest win rate. Sometimes your seat will put you into Gruul, and your best bet is to make the best of it. Data won’t replace your draft fundamentals!
Individual Card Evaluation
Speaking of fundamentals, I want to talk about single card evaluation. When going over cards I mostly look at GIH WR, and pay some attention to OH WR and IWD to get a sense of how much better the card is in your opener, and how key it might be to your deck (either by being a bomb or a big synergy piece). From there, I find the simplest way to categorize cards is into “good cards” and “bad cards” (unheard of, I know) and in turn, “Good” and “Bad” each are split into two sub-categories. Here, I’ll go over all four.
The card is ‘Bad’:
- The stats are bad on the card, and you think the card is probably bad.
These cards are bad, and there’s little you can do to make them good. Most cards with poor stats fit in here, and if you’re unsure about a card with bad stats, I’d stick it here until you see it in action. Heirloom Mirror fits into this category, as does Wake to Slaughter, and – of course – the most lovable card of the set: Stuffed Bear.
The card is ‘Sneaky Good’:
- The stats are bad on the card, and you think the card might be good.
This category is the biggest place where your skill as a player matters when it comes to overriding the data. These are generally cards that fit a very specific niche, but aren’t being played in the right decks. This set’s best example is Otherworldly Gaze. It can be a great card – but only if you have the deck to support it. If you put it in your zombie based Dimir deck, you’re doing yourself a disservice. Have a busted Azorius deck with 16 Disturb creatures and an Ominous Roost? Congrats, you’ve picked up a copy of Ancestral Recall (with flashback).
The card is ‘Good’:
- The stats are good on the card, and you think the card might be good.
I’m sure you’ve heard of Organ Hoarder by now, and I don’t think anyone is going to say that the card is bad. Hoarder is the prime example of a ‘Good’ card since it fits naturally into any deck, and you’re never really unhappy to see it. Like the ‘Bad’ cards, if you’re unsure about a card, and the stats seem promising, it should go here and you can re-evaluate it later. Chances are you won’t be disappointed. If a card was only good in some archetypes and not others, it’d probably have worse stats, and fall into the ‘Secretly Good’ pile.
The card is ‘Still better than you think’:
- The stats are good on the card, and you think the card might not actually be good.
I think this is where the fact that the data comes from a variety of skill levels really shines. Since you’re not always dealing with the best of the best, if a card is doing well, chances are it is just good. There’s basically no chance that a card is only doing well because everyone is playing it in the perfect build in the perfect way. Does that mean you should always put the card in your deck? Probably not. Does it mean the card is powerful? Almost definitely.
There is no ‘Sneaky Bad’
Going back to the breadth of the data, I want to close out on the idea that the data can hide cards that are ‘Sneaky Good’, but it can’t hide ‘Sneaky Bad’ cards. If you have a card that you think is ‘Sneaky Bad’, it’s important to think about why a card outperforms your expectations. If it was actually bad, people would still be playing it, but they’d be losing with it, which would make its stats worse. Given that, I find it especially useful to come up with some specific reasons which explain why a card does well, and then see if you can back those up. I’m sure I’ll draw ire for adding to the Lunarch Veteran debate, but I believe it’s the best white common, and the perfect example for how a card can’t be ‘Sneaky Bad’. If you’re down on a card with good stats, it might be overhyped, but it’s still just good.
Going back to Kvartek’s article, he compares Veteran to Hoarder and Adeline using OH WR. That’s not my preferred stat to use, as it biases towards cheaper cards and more aggressive strategies. And Veteran, by virtue of it being cheap and it being best when you play it turn one, has a high OH WR showing that it is good early. However, if you use GIH WR, a stat which is more long-game stat, Veteran’s stats are still great. Overall, Veteran has a 60.9% GIH WR, where Adeline has 65.6% and Hoarder has 62.1%. That’s still surprisingly close for a common and a bomb rare, even without the bias from OH WR. So, no, it’s not better than Adeline, but it might be closer to Organ Hoarder than you think. Finally and importantly, since the GIH WR is also high, it points to Veteran also being good when you draw it later. So, why might the Veteran be so good?
The first hypothetical reason is Veteran is a versatile card. It fits into any of the archetypes reasonably well. In UW it plays with the Disturb plan; in WB it can be sacrificed twice; in WG it helps to enable coven; and in WR it’s a 1-drop which turns into a flyer later, letting you be aggressive. When you look at the stats by deck colour and sort by GIH WR, Veteran shows in the top-3 for each colour pair with white, which backs up the idea that it’s flexible. If it only fit into one deck, it’s stats would likely be lower overall, and it wouldn’t show up as a top common for a colour pair.
A second reason that can explain why Veteran has such high stats is the life it gains over time. Raw life gain isn’t very good in draft, but the incidental life gain (that you’re getting as you build your board up) is incredibly relevant defensively. MID is a format where you can die even from double-digit life totals, and Veteran makes large alpha strikes harder for your opponent, and ping damage less threatening. That life can also be leveraged offensively, letting you be more aggressive. If you play veteran on turn one, and your opponent doesn’t remove it, you’re effectively starting with 5 extra life. When your creatures are swinging past each other, that counts for a lot. Unfortunately, I don’t have data to confirm or disprove that idea, but I’ll let you think about the games where you played/faced a Veteran.
I’ve blathered on for a while about stats and cards now, but, if you can find the delicate balance of when to follow the data and when to trust your own judgement, you’ll be at your best. You’ll have an idea of the best cards and easiest archetypes to draft, and you’ll also be able to gain an edge from the gems hidden below the surface of the data – the hard to draft archetypes, and the secretly good cards.
But above all else, my main point is this: You should trust the data more than you might think. Don’t follow it blindly, and definitely don’t use it to proclaim absolutes. But, enough people and enough games are in there that the data is going to be more right than wrong.