Probability in MTG: The Strong Law of Large Numbers and Racing the Meta


With Ravnica Allegiance right around the corner, the changes that its release will impose upon the meta is likely on many people’s minds. In this article, I briefly explain the Strong Law of Large Numbers (SLLN) and apply this to the challenge of racing the meta which is, at least for me, the main goal of playing MTG.

Strong Law of Large Numbers?

The SLLN is a theorem that states that if one conducts the same experiment over and over again, such that each trial is independent of the others, then the sample average of the outcomes will approach the true expected value of the experiment almost surely as one performs infinitely more trials. In the context of experiments where there are only two possible outcomes, particularly Deck A winning against Deck B when piloted by the same pair of players in each game, each player being a master in both decks and completely familiar with the other’s playstyle such that each succeeding game is not affected by any preceding games, then as the players play infinitely many games, the true win rate of Deck A versus Deck B when both are piloted without misplay will be obtained.

If it was possible to conduct this experiment for each pair of decks in a given card pool, then one can obtain all of the true win rates of each deck both versus a specific deck and versus the entire collection of possible decks. Of course, the reality is that it is not reasonably possible to conduct this experiment. For one, I do not think there are existing A.I.s that can sufficiently mimic the playing ability of  a pro or even a highly competent MTG players across multiple decks. However, those win rates do exist and are basically hidden information from players at the release of a new set. The reality is that the exact value of those numbers likely remain hidden indefinitely. However, estimates of those win rates can be ascertained as more games under the new card pool are played across different platforms and data from those platforms are made available. This is where the concept of racing the meta comes in.

Racing the Meta

As previously mentioned, while the ideal setup for finding true win rates cannot be reasonably implemented, imperfect experiments pertaining to a new card pool begin to be conducted upon the new set’s release. Each match played with the new card pool is one of those imperfect experiments and a large portion of these are documented and made publicly available through channels such as  MTG Goldfish. While SLLN cannot be invoked on this due to violations of its assumptions, the idea that estimates to the true win rates of each deck become better and better as more matches are played remains dependable.

Racing the meta means being able to identify which decks are best as early as possible, before the truth about those decks’ superior win rates become common knowledge. By doing so, one is able to play more matches using the superior deck, and therefore generate more value per unit time while the new card pool is standard legal. An important thing to note here is that overall win rates always depend on the existing meta. That is, incoming information about viable decks in the new card pool from match results across platforms affects what people decide to play, which in turn shape the new information generated. Thus, racing the meta also involves recognizing when to change decks as the meta changes.

For Guilds of Ravnica, I was able to achieve considerable success at the beginning of the season with a UB deck that only had The Eldest Reborn and a single copy of Josu Vess as win conditions during the first two weeks upon release (I had 4 Thief of Sanity in the sideboard). As more information about the new card pool emerged, it became apparent that UB was disadvantaged versus both Red aggro and Golgari with multiple copies of Midnight Reaper, and so I shifted to the latter.

It is also possible for a deck to sustain its high win rate across set releases. An example of this is Ramunap Red, which had obscenely high win rates versus the field the whole time that its entirety was legal in standard (Ixalan just made it stronger with Rampaging Ferocidon). Yet this was not obvious enough until WotC published the numbers which it used to justify banning cards from the deck. Imagine playing Ramunap Red from the time that Hour of Devastation was released up to the time when Ramunap Ruins and Rampaging Ferocidon were banned from standard. You would have generated X% more value (whether in gold, gems, tix, or $) than you would have playing anything else for the same number of matches. For me, it is this that constitutes a win in MTG and what makes the game so exciting. The game does not begin at the start of your match; the game is a race that begins when a new set is released and restarts with the release of the next. “Winning” is measured by how early you are able to figure out what decks work, through a combination of insight in evaluating the new card pool and empirical information on deck winnability that is released to the public. As more information is released, it becomes easier to identify what you should be playing. However at the same time, the value of this information is diminished because of the time you lost not figuring out the best decks earlier (during which you were playing other decks and were therefore not winning as often as you would have been).

Start your engines…

As we welcome the Orzhov, Gruul, Rakdos, Simic, and Azorious, I hope that this perspective can help you in making deckbuilding decisions in the new season. Put on your thinking caps and most importantly, never neglect to open your eyes and ears to the data about the new meta. Let the race begin…

…and may the shuffler be with you.

Comments