I've recently explained the concept of Elo, which is a ranking system which scores teams based on their performance against teams of varying skill. I took the Elo of each qualifier team, then calculated the historical frequency of victory for similar matchups to each match. I considered two factors:

About what is the rank of each team?

When calculating the historical frequency of similar matches, I only considered matches where both teams varied from their equivalent in the prospective match by no more than 100.

About what is the gap between the teams?

Then I only considered games where the Elo gap was within 100 of the Elo gap for the matchup I am projecting.

Easy enough, but that's just the odds of one game, while the group stages are running in two-game series. The odds of winning two games in a row are different than the odds of winning one game, and they are slightly more in favor of the underdog. A binomial function run once for each possibility will give us the likelihood of each end result.

What you end up with is the expected odds for every matchup for each possible option: 2-0, 1-1, 0-2. What you'll see in the chart below gives the most likely outcome in number of games (with the team on the left as the winner and the team on the top as the loser) along the approximate expectation for that result.

Limitations of these Projections

First of all, some teams have been very inactive leading up to the qualifiers (Leviathan comes to mind). This will mean that they could be misrepresented by their Elo--especially if they have been training. I checked the playhistory of every team for two things:

How much have they been playing

Teams that don't have a lot of recent games don't have very reliable rankings. Teams that lose unexpected matches against unreliably ranked teams will end up at a serious disadvantage out of the group stage.

Whom have they been playing?

If teams have been playing a lot of higher-ranked squads (for example, if North American Rejects had been playing against EG, Secret, and Empire for most of their games) over time they would probably lose a lot of Elo relative to their peers who are only playing against similarly-ranked teams. So I checked to see how much Elo was being lose in small increments (e.g. against much better teams) and found that no team had their Elo significantly altered by this factor.

In order to prevent the possibility of a very recent surge or drop of Elo from one or two games, I used averaged each team's Elo over the last week rather than just using the current Elo, which will increase the reliability of calculations between a team that took the last week off and a team that has been very active over the last week.

Other limitations

I used a historical model based in 6.83 winrates. This will not be 100% accurate in 6.84, but it should be very close. If you want to know why I am confident that Elo remains approximately equivalently predictive across patches, see my most recent Metagame Fortnight.

Some teams are better at taking upset victories than others, while some teams are better at winning against lower-ranked opponents than others.

My model does not, unfortunately, take into account the type of victories individual teams are best at achieving.

This is something I'm looking to incorporate in the future, but for now some squads may be under or overrepresented because their individual ability against SPECIFIC opponents is not taken into account.

I cannot stress enough that

statistics are not fortune tellers

. It would actually be

very weird

if ANY of these group stages came into place exactly as shown, especially since so many individual matchups are projected at less than 50% certainty.

One of the beautiful aspects of competition is the desire to overcome the odds.

No team has a 0% chance of winning, and no team has a 100% chance.