Magic: the Gathering is currently at a particular point in its development. On the one hand, it is The Original Game of its kind. It has a vast tradition, a player base with decades of individual experience, and a very well-developed game strategy base. On the other hand, the information we get on the formats is very qualitative and, in large part, based on punditry. Formats are “fast” or “slow” – but what does it mean?

Most pundits/podcasters use simple terms and, don’t get me wrong, are entirely accurate in their assessments. Still, with the rise of online play and tracking, I think we are ready to put some numbers on the format speeds, on the mana costs of successful decks, on the propensity to mulligan, and most importantly, link those values with the ultimate measure of success – the win rate. Data can also help us where the pundits’ experience is lacking: assessment of novel mechanisms the game presents us with, like the very much hyped Modal Double Faced Cards (MDFCs) from Zendikar Rising (ZNR).

Cragcrown Pathway // Timbercrown PathwayRiverglide Pathway // Lavaglide Pathway

This article will try to ask two questions: how to measure the format speed (using Best-of-1 drafts from Magic Arena) and how does the speed of the ZNR impact the usefulness of the MDFCs. The large datasets also allow us to look through many more games of Magic than any pundit can play to draw any conclusions. To give you the impression, ZNR dataset I look at, has over 25 000 games, taken from roughly 4 000 drafts by 750 players.

Speed of the format

The first topic I wanted to address is the speed of the format. And straight down, I ran into a problem: What is a useful metric of the speed of the format? Is it how fast the games end? Is it how cheap the decks are? Is it how much advantage does being on the play you get in a format? There is not a trivial answer to those questions. Think about this: which was a faster format: THB or M21? I assume most of you will answer M21. But the average CMC in a deck was lower in THB. Several factors can impact an average casting cost in a format – think cycling decks with ten one-mana cyclers that might have a CMC of 6 or kicker spells that you hope to play as 6 CMC cards on paper are 2-drops in ZNR. But that doesn’t mean you will not curve out with ease, and you will not sit with your uncastable cards until turn 6. That is why your deck’s CMC will not tell the full story of how fast or grindy it is (Figure 1).

(fig.1)

If you look at how many turns do the games last, the differences are relatively (and to me, very surprisingly) small. In all of the last formats, 60% fastest games ended on turn 10 (Figure 2A). The average turn a game ends varied from 10.2 turns in THB to 9.2 turns in M21 (Figure 2B). Looking at the distribution of when the games end, you generally see little difference. But those small differences do matter, even if they are hard to capture by data.

(fig.2A)

(fig.2B)

I looked at multiple other metrics to try and find one that captures the speed of each format. These observations made me conclude that the format speed’s best measure is the difference in win rate on the play and on the draw (Figure 3). In very aggressive formats, being on the play is essential. If you get an earlier board presence, you will be one step ahead of your opponent, and the additional card they get from being on the draw matters much less. Fast formats are fast not because they have cheap spells, but because of any early advantage, one of the players gains snowballs very quickly. This means, quite intuitively, you are more likely to end the game before the opponent can efficiently utilise any extra cards they drew.

In slower formats, on the other hand, you will have more opportunities to stabilise. You will be more likely to implement a grindy game plan, and the extra card you get is relatively more important than the advantage you get from playing your lands first. Alternatively in prince formats, all the game before the turn one of the players plays the unbeatable bomb are neutralised.

(fig.3)

The pretty sophisticated distribution of the game durations tells us little about the format speed. Therefore, it was quite surprising that the much less fancy metric, the average length of a game, correlates pretty well with the difference in win rate on the play and on the draw (Figure 4). It will be interesting to see if this is a universal rule with the data from future formats, but so far, there is a very linear connection between the game duration and the advantage of being on the play. With the data available, it, at least, looks like an exciting correlation that could potentially help us assessing the speed of the format by proxy just looking at how beneficial being on the play is.

(fig.4)

Enter the MDFCs

According to the data, ZNR is a deceptively fast format. Most analysis of the format so far put it as medium to slow, but the actual average game end, plus the advantage of being on the play in BO1, puts it as one of the fastest draft formats of last years, slower only than M21 in terms of game duration and only M21 and ELD promoted the player on the play more. This made me think: is the drawback of entering the game tapped in its land mode larger than we originally anticipated? Is it possible that it outweighs the undeniable power of it serving as an additional spell later in the game?

To try and answer those questions, I looked at three metrics. The first of them was: how many lands should you play? This is a very nontrivial question. Should the MDFCs be treated just like any land, or maybe should they be counted as spells to some extent? If the first is true, lower land counts should yield higher win rates. If the second is correct, you should see the reverse. The answer was pretty straightforward; decks with lower land counts (land counts do include MDFCs in their count) do better on average (Figure 5). But that is a little surprising. MDFCs are hyped to be very powerful because they let you play more lands with a relatively small risk of flooding. But putting more lands seems to decrease the win rates.

The reason might be the situational character of many spells linked to the MDFC lands. A card like Beyeen Veil can be handy in some situations, but it is just not doing much in many other board states. If you play 21 lands, this card replaces a playable card, which means a good spell in modern limited formats. If it is useless with 20+ lands in the deck, it will most likely be yet another land you play as you flood out, providing a minimal advantage. The same card in a 17-land deck will be relatively more useful as a land, as you have fewer of them, and when it is favourable to play it as a spell, it will very likely be worth more than the missed land drop.

(fig.5)

But the number of land does not tell us anything about the number of MDFCs in those decks, although on average, decks with fewer lands also have fewer MDFCs. If MDFCs are unequivocally powerful, decks that contain more of them should, on average, do better than those without them. And the difference should increase with the number of MDFCs. If they are detrimental to the deck power, you should see the reverse. The most likely scenario, though, is that there is a broad optimum of MDFCs – you want some in your deck, but if you overdo them, it will hurt your win rates.

Here the data is much messier (Figure 6): there seems to be a slight drop-off in win rate if your deck has more than 6 MDFCs, but the differences between having 0 or 6 MDFCs look non-significant, nor is there a clear trend. But this data is confounded by several factors. It does not take into account how many lands are there in total in a deck, which means that on the higher end of the spectrum, you have a mixture of 17-18 land decks and, at least according to previous data, poorly constructed 21-land decks. To compare likes with likes, I also looked at only the decks with 17 lands (as they had a good mix of 0 to 5 MDFCs, Figure 7) and their win rates based on the no. of MDFCs. But the picture did not change dramatically. It does look like 17-land decks with 1-4 MDFCs do similarly well, but anything over 4 is detrimental. It does not look like, though, not having any MDFCs does not damage your deck’s performance significantly.

(fig.6)

(fig.7)

The last (I promise) analysis of the MDFC performance has something to do with their less discussed drawback. In any Zendikar Rising material I encountered, the fact that they come into play tapped was swept aside without much afterthought. But this, in a fast format, can be a significant drawback. To see if that is the case, I looked at the composition of opening hands. In BO1, the vast majority of opening hands have 2, 3, or 4 lands. If a land entering the battlefield tapped is a problem, hands with the same number of lands that contain MDFCs should do relatively worse than hands with only basic lands. The more lands you have in your opening hands, the less of a problem should it become.

That is precisely what I saw in the data (Figure 8). The two and 3-land opening hands where some of the lands are MDFCs do worse than those with exclusively untapped lands. But if your opener has four lands, this difference not only disappears but is turned around (albeit by a tiny margin). This means that MDFCs entering into play tapped is a real cost. And that it needs to be proper factoring into the cost-benefit analysis of the MDFCs. It is worth noting that the cost may not be a problem for all players. It is very possible that misplays of lands and bad decisions early in the game by a smaller subset of players can be the reason for this discrepancy.

(fig.8)

To conclude: each format’s speed can be characterised by the ability to get in front and stay in front. My gut feeling is that the princely formats, where bombs matter, will be more resilient to such snowballing effect. In pauper style formats, where commons and uncommons dictate a deck’s power, it will be harder to turn a game around once one side gains advantage, benefitting being on the play. Based on these measures, ZNR is a relatively fast format. This affects how powerful MDFCs are. You do not want to replace your playables with MDFCs, and you should try to keep your land count within the traditional 16-18 lands. Secondly, you probably want to keep your MDFCs to not much more than four. Thirdly, the cost of MDFCs entering the battlefield tapped is real and should not be dismissed lightly.

Some disclaimers: The data was very kindly given to me by Robert Conroy, running 17lands tracker. Open data is something not only useful but also valuable. We should all be grateful that Robert and others involved in 17lands are willing to share the fruit of their hard labour as it benefits the whole community. Secondly, the data comes from a non-representative group of players. The average win rate of 17lands user is >50%. This means the problems I describe here affect better than average players. Thirdly – the data shown comes from BO1 matches. These, thanks to the hand smoothing algorithm, have much less variance in lands in the starting hand. This means that, compared to BO3, the benefits of MDFCs can be diminished, as thanks to the algorithm, you will have fewer hands with <2 lands, a situation where expanding your mana base helps, and fewer hands with >4 lands, where the spell part of MDFCs lets you mitigate the flood. With more games in the format, the same analysis for the BO3 would help draw conclusions relevant to all BO3 formats on Arena, MTGO, and paper. Lastly: this article’s goal is to start a discussion about MDFCs from a fresh, less rose-tinted perspective. But the data presented should not be treated as the ultimate authority. Magic data, especially Limited related is notoriously noisy, there are many confounding factors. MDFCs are a fascinating topic and it would be interesting to figure out how to use them properly. In the current situation to do so both data and punditry are extremely useful, if only to feed each other with ideas on how to solve the problem.