What did we learn from a thrilling play-in tournament? For one, predictions are harder than they look, especially at the start of a new meta. Last week, Jonathan “Reinforce” Larsson’s seasoned analytical mind just won out over stats producer Ben “CaptainPlanet” Trautman’s SOLOQ system (final score: 2-1), but they’re back for an even more daunting task—predicting the entire double-elimination bracket. Let the clash continue!

The Statistician: CaptainPlanet

Yeah, so I lost to Johnny. What of it? I’ve learned my lesson and have come back prepared for the real game: predicting the outcome of the double-elimination bracket. Reinforce can have his kiddie pool of a play-in prediction win, I’m gunning for the big prize.

I wrote in last week’s article that my two systems, SOLOQ and OLE (which SOLOQ is built off of), had one major weakness: they’re not very predictive at the beginning of a patch. While they provide a great framework for carving weighted paths through future match outcomes, all of that doesn’t matter if I don’t have a decent idea of each team’s true strength entering the competition.

Given the results of my play-in tournament predictions, we can see how the model failed: it put too much emphasis on Stage 4 performances, which led it to predict that the Guangzhou Charge would overpower the Chengdu Hunters and Seoul Dynasty and that the Philadelphia Fusion would have beaten the London Spitfire, had they even reached them in the first place. Instead, the Charge got rookie jitters against a Dynasty team that had clearly game-planned for them, and the Fusion had a bad initial take on the meta.

Therefore, for this much-higher-stakes competition, I’ve performed some experimentation. In order to improve the accuracy of SOLOQ’s initial team strength assessments, I implemented a new system. This system involves the Sourcing of Credible Rumors and Information from Many Brilliant Up-to-date eXperts: SCRIMBUX for short. I use these SCRIMBUX to make a SCRIMBUX correction to the SOLOQ inputs according to a specific set of rules.

Here’s how it works: I polled as many Overwatch League coaching staffs who were willing to participate in this endeavor for their own personal power rankings of playoff teams (besides their own) on the playoff patch, 1.39. I combined these rankings, and then compared them to the rankings of their Elos entering the season playoffs. Here is that comparison below*:

Team Elo entering playoffs Rank SCRIMBUX rank
San Francisco Shock 1,155.3 1 1
Atlanta Reign 1,105.6 2 7
Seoul Dynasty 1,086.3 3 6
Hangzhou Spark 1,075.6 4 8
Vancouver Titans 1,029.9 5 2
Los Angeles Gladiators 1,015.4 6 5
New York Excelsior 980.9 7 4
London Spitfire 960.7 8 3

*The Spitfire and Dynasty’s playoff Elo are post-play-ins, while the other six teams have the typical 30% reduction towards the mean on a new patch in OLE.

The next step in the SCRIMBUX correction was to take 25% of each team’s Elo and put it into a shared pool. This pool was then redistributed according to the difference between each teams’ SCRIMBUX rank and their initial playoff Elo rank. The most I felt comfortable augmenting this redistribution was by about 30%, so I mapped the difference between their playoff Elo rank and their SCRIMBUX rank to a percentage equivalent to five times that difference. This created a range from 0% (Shock) to a max of +/- 35% (which no team hit anyway).

To calculate the redistribution, each team received back a share of the Elo pool equivalent to the average value of the Elo donated—262.8—plus or minus the above percentage. For example, the Shock donated 25% of their Elo (288.8), and received back 262.8, since their SCRIMBUX rank is equivalent to their initial rank. They lost a little bit of Elo because we are still somewhat uncertain of just how good they’ll be on 1.39, even if they’re the consensus favorite. As a different example, the Spark donated 25% of their Elo (268.9), but only received back 210.24, or 262.8*0.8, as their SCRIMBUX rank is -4 from their initial Elo rank.

The SCRIMBUX correction has the effect of slightly flattening out the Elo values of the playoff teams closer to their group average while also re-shuffling their discrete rankings into (hopefully) a more accurate state—while still maintaining an equal amount of Elo rating in the system. Below, I’ve posted the post-SCRIMBUX correction ratings that were ultimately fed into SOLOQ to simulate the outcomes of the playoffs:

Team SCRIMBUX correction Elo
San Francisco Shock 1,129.3
Vancouver Titans 1,074.7
London Spitfire 1,049.0
Seoul Dynasty 1,038.1
New York Excelsior 1,037.9
Los Angeles Gladiators 1,037.5
Atlanta Reign 1,026.3
Hangzhou Spark 1,016.9

And finally, the Reinforce-destroyer—my bracket, as predicted by SOLOQ:

Here’s one interesting thing to note about this bracket, however. Even with all of this pre-work to make sure that my inputs to the SOLOQ model were as accurate as possible, this is a long tournament. By the end of the playoffs, 14 matches will have been played, and there are many, many different paths to glory...or failure. In fact, out of 100,000 iterations, 15,763 unique paths through the bracket were created.

The bracket above is the most common path that was predicted by SOLOQ, but it only occurred in 37 out of 100,000 simulations, a measly 0.37% of the total. By comparison, the model is much more sure about who will ultimately win the title (the Shock, in 24.5% of all simulated outcomes) and who will be the runner up if the Shock win it all (the Titans, in 21.8% of all Shock wins) than it is about how exactly each team actually ended up in the finals. Buckle up, we’re in for a wild ride.

The Analyst: Reinforce

The play-in tournament and the introduction of Sigma was an incredibly eye-opening experience for what’s to come moving into the season playoffs. With the Dynasty successfully shutting down Charlie “Nero” Zwarg’s Pharah and consequently the hopes of a miracle run from Guangzhou, they beat the odds and qualified for the double-elimination bracket, where they’ll first go up against the Vancouver Titans.

Meanwhile, London displayed multiple compositions that included everything from Jun-Young “Profit” Park on McCree, Ji-Hyeok “Birdring” Kim also on McCree, Birdring on Sombra, and Profit on Mei... all in what seems to be a Reaper and Doomfist meta. Strategical prowess, or sheer panic? I’m honestly leaning towards the later after reviewing their decision-making. Perhaps “scrimbucks” aren’t worth cashing in on after all.

It’s become clear to me, after reviewing the play-in matches, that every default composition starts with a Doomfist and Reaper composition. Then, based on the map layout or player strength, you mix in Pharah, McCree, and Mei to counter each other accordingly. Based on this presupposition, I’ve put weight in my predictions behind player strength on those heroes, historical ability to adapt according to the opponent, and pure intuitive decision-making rather than tactical execution.

With all that in mind, here is my bracket:

Hangzhou, a team I originally expected to do well, have dropped in ranking following question marks surrounding their Doomfist play, leaving me to think perhaps we’ll see more Pharah play coming out of them, a hero that Chung-Hee “Stitch” Lee and Byung-Sun “Fleta” Kim from the Vancouver Titans and Seoul Dynasty, respectively, would be suited to deal with.

The Los Angeles Gladiators enter my top four after believing in Gui-Un “Decay” Jang and Lane “Surefour” Roberts’ ability to handle the many adaptations required and their overall damage prowess. Perhaps relying less on their insufficiencies in the tank line—after Chang-Hoon “Roar” Gye and Jun-Woo “Void” Kang struggled in Stage 4—will be a good change for the LA squad.

But still, the San Francisco Shock and Vancouver Titans remain the two best teams in the league in my mind.

With the best coaching staff in the league mentoring Jay “Sinatraa” Won’s Doomfist and perfecting the team’s overall tactical execution, San Francisco will continue to punish unforced errors from lesser teams. And with a regular-season record that speaks for itself, Hyo-Jong “Haksal” Kim’s strength as a Doomfist and Pharah, and an overall roster core and team synergy that just screams championship-caliber—the Titans will beat most teams with their impeccable ability and ultimate usage.

But in the end, none of that really matters. We’ve seen it time and time again. Miracle runs through APEX brackets and stage playoffs that feel borderline scripted and eerily magical. Overtime situations that go from unwinnable to flipped on its head in a matter of a few heartbeats. The Titans and Shock are just a different breed of players. They’re champions. And come playoffs they’ll show us that even when stacked up against the odds, when that overtime wick starts burning down, they’re just a little bit crisper, a little bit cleaner, and they’ll overcome any obstacle you put in front of them. We’re in for one hell of a playoffs—it’s time to watch them go to work.