Giving this its own post.
Re-Sampling Link
https://docs.google.com/spreadsheets/d/1QpZ0IccGk5SoY54TJBeTZdvdmJ_ysPVKrQLWAHjAggM/edit?usp=sharing
We can simulate the results of the event from a single-elimination perspective (i.e. the statistically likely 1-0s, 2-0s, 3-0s etc) by calculating win-rates conditional on the actual metagame of decks remaining in a given round and decks remaining in each round based on past win-rates. Given a starting proportion of decks, and matchup win-rates (i.e. Shops beats Gush 60% of the time or whatever), we can calculate a round-by-round evolution of a metagame. That is what the linked spreadsheet does.
Choosing 5-0 (round 6 representation) as a cutoff, we get a predicted metagame of:
18.85% Shops
21.45% Gush
26.78% Dredge
8.76% Big Blue
1.93% Blue Control
11.99% Combo
5.07% Oath
4.70% Eldrazi
0.47% Other
This would predict a Top 8 of something like
2 Dredge
1-2 Shops
1-2 Gush
1 Combo
0-1 Big Blue
0-1 Oath or Eldrazi
Note that this is not supposed to be a "solved" metagame. This is the expectation if we ran the same tournament again with the same decks and the same matchup-vs-matchup winrates.
I think this is a pretty good facsimile of the 5-2 or better results.
The actual 5-2 or better results by archetype:
20% Shops (3)
13.33% Dredge (2)
26.67% Gush (4)
20% Combo (3)
6.67% Big Blue (1)
6.67% Blue Control (1)
6.67% Eldrazi (1)
The deviation between the actual 5-2 results and the predicted 5-0 share can largely be attributed to pairings. There is some methodological weakness (i.e. single elimination as opposed to roughly triple elimination with brackets based on current record) but this is likely overshadowed by the effect of pairings. Dredge, based on this, got extremely poor pairings. (i.e. average pairings with exactly the same matchup performance would have resulted in an even stronger Dredge showing) Oath also appears to have unfavorable pairings. Combo decks were the biggest winners in the pairings with a 5-2+ share that far exceeds their predicted 5-0 share. Gush somewhat over-performed their expected result.
Gush's win-rate: For the case where no one drops out of the event, the above figure of 46% is correct. If players drop out as they lose by the 5th round Gush will have experienced a win-rate of 45.73% (i.e. 5-0s and only 5-0s in contention) and by the 7th round a win-rate of 45.70% (i.e. pure single elimination). So the tournament environment definitely gets more hostile to Gush as the rounds progress, but not quite to the level of 43% overall win-rate (the actualized result). This means that Gush pilots got somewhat unlucky to experience their overall result this weekend. Nonetheless, I cannot construct a sampling method that gives Gush a >50% win-rate in the environment of this particular tournament.
One thing worth noting is that due to the nature of Swiss tournaments any given deck will experience the greatest number of live-for-Top-8 rounds at the beginning of the tournament. The pool shrinks exponentially, since every match has a winner and a loser each round eliminates on average half of the contenders from Top 8. To estimate the odds of winning the event or placing very highly you want something that weights the last few rounds as heavily as the early rounds. I don't really want to do more spreadsheet math right now but I would point out Gush's expected win-rate late in the tournament is down in the 43-44% range.
Given that Gush over-performed in the 5-2 region and under-performed overall I believe the only reasonable conclusion is that Gush had a higher variance tournament than other decks. The data support the conclusion that Gush pilots had a wider variety of player skill than the average deck overall.
