A statistical look at the Standard metagame

So I've been tinkering a little bit with an application that takes all the Standard decks played in Magic-League tournaments and determines what are the common decks being played by finding the similarity between all pairs of decks and then clustering the resulting graph. The clustering is fairly naive; I just consider the clusters formed by the subset of edges with the greatest weight (similarity).

Similarity is the percentage of different non-land cards among two decks, in a nutshell. I only looked at the clusters that represented at least 1% of the decks (I had about 4200 decks downloaded), and these are the best results I got, considering pairs of decks that are at least 80% similar:

541 Faeries
465 Five Color Control
313 Kithkin
281 Red Deck Wins
203 BG Rock
160 Reveillark
141 Merfolk
52 Doran
49 Ten Commandments
2205 decks covered

Note that this doesn't cover all of the decks played, since there are a lot of homebrews, variants, and plain ol' bad decks played. But this is an interesting look, considering that on Magic League there is no issue of card availability.

In the future, I'll expose this via a web tool. Some other ideas are to break down variants of each large archetype, break down matchup statistics (since Magic-League publishes all match results to their tournaments), determine the most characteristic decklists for an archetype (using some kind of node centrality metric), and implement some kind of tagging system to name archetypes.

If you have any other ideas for this, I'd love to here it!