Deck Archetypes are Hurting MTG Metagame Reporting
MTG metagame reporting has been basically stale since a decade ago, when many players (including myself) started recording metagame info via google forms and reverse engineering the winrates with the public pairings.
Competitive magic was very different back then: MTG Arena didn't exist, and neither did melee.gg, where most tournaments are played nowadays. But the way we report the results is basically the same.
Last metagame report by Frank Karsten
Modern metagame of Summer 2018, compiled by me
We are basically displaying the same info (plus the nice confidence intervals), but nowadays we have much more information than back then. Then we had to settle for archetype, since we didn't have access to the player decklists. But we don't need to do that now, we have full decklists of everyone!
But what is an archetype anyway?
Macro archetypes are core to the way we understand the game (and have been for decades now). From the basic aggro, control and combo, to the more nuanced midrange, tempo or ramp. But since we want to distinguish between two decks that share the same strategy, we name those decks. For example: Mono-Red Aggro or Gruul Mice. We consider those different archetypes, but how different are they really?
In this example they share 48/75 (64%) cards, including the high power trio of Heartfire Hero, Manifold Mouse and Monstrous Rage. For the purposes of reporting, these are the same deck. For the purposes of playing, you block a double striking mouse and get monstrous raged out of the game in both cases.
The metagame diversity trick
In recent years we have had some pretty homogeneous metagames. Take for example PT March of the Machine. According to the data, 6 decks had roughly 5% or more metagame share, not so bad. But if you look more closely, 53,2% of all the decks played both Fable of the Mirror Breaker and Bloodtithe Harvester.
PT MOM metagame report. Credit: Wizards of the Coast
Was this a diverse metagame? Of course not, and little after Fable and others were banned. But I would argue that the picture is telling you otherwise.
Don't get me wrong, the deck metagame share picture is fine, we are simply overusing it. And we don't need to. We have better data now.
Lost in the archetype
A clear limitation of the current model of reporting is that it can't tell us anything about mirror matches. Izzet Prowess vs Izzet Prowees will always be 50% by definition. But what if I play the full 4 Drake Hatcher vs someone that plays none? Sorry, same archetype. Can't tell you anything.
This problem is also true with any particular version that has some different card choices but is classified as the same archetype. Is the Drake Hatcher version better against Mono-Red? With the matchup matrix we can't tell.
The Solution
I've been theory crafting something for quite some time, but I want it to be polish enough to not embarrass myself. But in essence, we should use the full 75 and not just the deck archetype (btw, most of the time this field is autoreported in melee.gg, and the people that generously do the stats for us manually curate it. Been there, done that. It sucks).
For example, card triplets (or any sequence of N cards) is a very simple way of getting how good different versions of the same deck are against each other or other decks. We could calculate the winrate of 4-Cori/4-Drake/4-Opt vs 4-Cori/0-Drake/4-Opt.
For metagame reporting we could look into biology or economics for some inspiration. But remembering that we shouldn't use the archetypes, but the deck lists.
TL;DR
- Archetypes are loosely defined and are a poor metric of metagame diversity and performance
- We are under using the data we have available
- Deck lists give a more nuanced view into the metagame