Grouchy About Groups
As discussed in myriad articles, tweets and online discussions over the last few years — the format of competitive events has a very significant impact on the outcome of that event.
Quite obviously the initial seeds and placement of teams into various formats can also have a significant impact in some types of formats, whereas in other formats it makes little to no difference (for example, in a big round-robin group stage). The current Major format to be used at Chongqing (4-group GSL into double elimination with top 8 in WB & bottom 8 in LB) is part of the former category, and it’s a format that’s been endorsed and requested by Valve.
There are now two tasks to discuss: how to seed teams, and how to effectively allocate teams into the four groups in the most efficient way. After discussing these, we can circle back to talking about qualification as a whole.
Allocating Teams Into Groups
One thing everyone (me included!) loves to do when groups are announced is discussing which is the ‘group of death’, ‘easy peasy group’ and so on.
In some situations having slightly “imbalanced” groups (as in, different average rank, or different average ‘skill’, or an unequal distribution of the top 8 teams) is actually the optimal method to ensure traditional seeds propagate into the playoffs (which are essentially Double Elimination brackets in which the first round has already been played). This is due to how pairing works going from groups into the playoffs: teams as a general rule cannot play against someone from their own group for as long as possible (assuming there are no upsets) in the playoffs. This creates a few limitations, but not enough to make this impossible.
We can easily work backwards from this playoff bracket to the groups. One thing to note is that the 4 groups need to be symmetrically allocated: each group needs to have a #1 seed and #2 seed, and have these seeds split such that one appears in the top half and one in the bottom half of each bracket (both in the winner’s and loser’s bracket). An easy solution to this is below:
This leads to the following final distribution, which fortunately provides groups which fulfill all the requirements above in addition to having equal average-rank teams (some groups are more top-heavy than others however).
Now onto the more contentious issue: seeding teams. Without rehashing previous discussions on the matter, I’ll start out by giving my ground rules: the method for seeding teams needs to be a purely objective, fair, and transparent rules-driven system which reasonably correlates with reality.
I’ve spent some time thinking about this, and created a system with only a few key rules. I proposed these rules to Starladder when their qualifiers were reaching the playoffs phase. I’ll provide an example after each rule (except the first, which is boring).
The first is that the Minor Winner has to be seed #16. This just makes sense from a logistics perspective: groups shouldn’t shuffle around a huge amount just based on the winner of the Minor one week prior to the associated Major. The team that wins the minor also has the least preparation time, and is the most fatigued.
Secondly, the most important datapoint reflecting “performance” or “skill” of a team that we can be sure about is a good result in the previous major. This is consistent with how Valve has basically always evaluated invites to their events, so it makes logical sense to enshrine that belief in the rules. Consequently, any team which comes top 8 at a Major and manages to qualify (except as a Minor Winner as per above) for the following Major is guaranteed at minimum that same seed.
After this, regionality is probably the next best proxy for skill, based on historic data, followed by their regional finish. This suggests that we need to evaluate “missing spots” and the regionality of the teams that should replace them. If a team which performed well in one major is absent at the following one — the most likely explanation is that similarly (or higher) skilled team(s) have edged them out (who obviously must be from their region). The logical solution is to consider those teams that’ve edged them out as being roughly similar (or slightly better) in overall (international) skill.
We’re now tasked with finding the highest unallocated EU team to take the 4th seed (in this case Team Liquid), then the next highest EU team + SA team + 2x NA teams to take slots 9–12 (i.e. Alliance, Chaos Esports, J.Storm & Forward Gaming), and the final remaining teams to take slots 13–16 (No Pango, Thunder Predator, Aster). This leaves us with the following:
The final step is sorting out ties between the recently allocated slots. The first tiebreaker is final result in the previous Major, followed by position in the regional qualifier. I’ve not thought too hard beyond these, but I’m sure there could be some reasonable objective tiebreakers from here on out (DPC points, head-to-head results, etc).
If you now take these seeds and apply them to the part #1 groups, you get the following effective groups:
There’s one related topic that’s been discussed quite a bit recently — the highly contentious seeding matches of the Major qualifiers. The biggest issue with these was the lack of communication around how they would impact seeding. Ultimately I think it’s unavoidable that some matches played in a qualifier could end up not actually impacting the seeds for the associated teams that much — for example if EG lost against Plus Ultra then effectively J.Storm vs Forward Gaming would’ve been for a top 4 seed, but because EG won that it made the J.Storm/Forward game less meaningful (still impacted their tiebreakers as per rule #4).
That said, the seeding matches are there to fill in the gaps should there be any such gaps, it’s better to have some limited data than none at all. To me this is an acceptable inefficiency in the DPC system as a whole — and a confirmation that LAN results against a wide array of teams (from multiple regions) is way more significant than an online regional qualifier; despite the obvious time effects (results from ~6–7 weeks ago versus 3–4 weeks ago).
I don’t believe that the method described above is infallible — I just think it’s just a good step towards coming up with a pretty clear and totally objective system to determine seeds, in a way where any fiddling done by a tournament organizer is transparent to all parties involved.