It’s been a minute since our last supermajor. Coming up this weekend, we have the return of Don’t Park On The Grass – a tournament series quite near and dear to my heart. This is in large part due to the event’s history with EndGameTV, a-now-defunct organization that’s also near and dear to my heart due to its history with many of my friends, as well as the multiple tournaments it ran. In recent times, EGTV has been effectively succeeded by Level 1 TV, a group whom I believe has already done amazing work for the scene and seems poised to take an even bigger step forward.
I can’t overstate the stacked nature of this bracket. Although Mang0’s not in attendance, basically the entire rest of the Top 10 is here. As a nice little treat, we have Leffen there too, though we’ll see how that shapes up by the time of the event. Even if he DQs, the rest of the 64 seeds have practically everyone else you could imagine: bonafide Top 25 players, Top 50 mainstays, and Top 100 hopefuls alike. Hell, I’ll be there too. If you see me and want to say hello, feel free to do so (though if you start yelling at me, it will be very unpleasant, so don’t do that).
Today’s column is all about DPOTG 2024. In similar fashion to what I did for Tipped Off 15, I’ve simulated the results of this tournament’s Top 64 seeds 1000 times. After explaining some brief changes to my methodology for this edition, I’ll jump right into what I learned from my simulations, most notably whom I think has the best chance of breaking out (or falling off) given the existing brackets here. Following that, I’ll briefly ‘summarize’ the groups of players here, talk about the limitations of my process, and then make my final pick for DPOTG champion.
Methodology
If you’re familiar with my methodology for predicting events, this column won’t be too different. Essentially, what I’ve done is recreate a Top 64 bracket through a spreadsheet format. Although this version of my tournament simulator has yet to incorporate an advanced losers bracket matchmaking format (due to the extremely complicated and timely process of accounting for double jeopardy), what I’ve done in the meantime is create stand-in 65th place matches, rather than setting a bye for losers of the first winners bracket matches. Waiting there, in the losers bracket, were 32 instances of a stand-in player I called “Top 96 Player.” These are the regional up-and-comers that bracket nerds love to talk about – everyone from the UK’s very own max to Utah’s OkayP. For the sake of time, I have effectively treated them as one player with several different instances.
Another key difference of my most recent approach is that I went back to manual win-rate setting, rather than using what I had previously set up for TIpped Off (a hybrid of Elo, head-to-head history, and matchup adjustments). This was because of two reasons: my initial win probability calculator was overfit for the first half of the year, and I didn’t have enough time to make proper adjustments for players who have risen in stock since. As a result, like I did with my LMMM preview, I used a mix of head-to-head history, ‘comparable’ head-to-head history, and consulting with my dear friend and When’s Melee reporter s-f. Together, we generated win probabilities for every possible matchup within the Top 64 seeds and also included matchups against our aforementioned “Top 96 Player.” On one hand, the numbers-based readers may roll their eyes at this process. In a sense, I get it. I have no illusions about this process ultimately fitting under speculative and subjective criteria.
However, I still think it’s useful. I ended up being satisfied with what we came up with together, and for most matchups within our Top 64 (technically 96), we had a pretty good structure for determining win probabilities. We essentially saw it in five categories (with gray areas between them):
- Tier 1: Needing a DQ/sandbagging/extenuating circumstances for an unexpected result (0.9/0.1). Example: Cody Schwab/aMSa.
- Tier 2: Strongly favorable/unfavorable (0.8/0.2). Example: Jmook/Fiction.
- Tier 3: Convincingly, though not dominantly, favorable/unfavorable (0.7/0.3). Example: Salt/SDJ.
- Tier 4: Competitive, though likely in one person’s favor/disfavor (0.6/0.4). Example: Zain/Jmook
- Tier 5: Coin flip or unknown (0.5 for both sides). Example: Hungrybox/Plup.
If you’re curious about the specific ratios we set, it will be linked in the appendices. We came up with our ratios through looking at every single Liquipedia head-to-head for 2024, checking for patterns across multiple head-to-heads, and seeing what has translated to results.Think of our ratios as our expected outcomes for how many times a player may win against another player in ten sets.
Seed vs. Average Simulated Placement
Those of you who remembered my Tipped Off preview will recall my picks for breakout players. In this segment, I’m going to do something similar and talk about the people attending this event whom I want to buy or sell. It’s nothing personal – I’m strictly talking about what the data says, according to the model I built off my starting points for set prediction. Wait. I guess that does make it a little bit personal. Still; you hopefully get the idea
NOTE: For this graph and other graphs I generated with Python, I ran into an issue with “null” coming up as an input value. While there are several ways around it, I thought it’d be way funnier and easier to title him as “Gio.”
Some of you are wondering why Cody and Zain are above the trend-line. In similar fashion, aMSa, Plup, and Hungrybox are above it too. But technically speaking, as very high seeds, their average simulated placement is beneath it. At the same time, for the purpose of looking at most players who aren’t going to be winning the tournament, this should prove to be a useful graph.
Breakout Picks
Ironically, there’s nowhere better to start for potential breakouts than Leffen, the biggest wild card of the whole event. His spot here is appropriately fitting: a plucky 11 seed that could either flame out in embarrassing fashion or destroy everyone like it’s Evo 2018. Other breakout picks here include some of the Top 48 to 64 seeds: each of Frenzy, Dawson, Faust, null, CPU0, Zeo, and Kacey stand out as great choices to take their results to the next level. If I had to pick two of them, I’d go with Kacey, a player that’s beaten Zamu, Panda, Dawson, and practically scored a win over a Top 100 player at each of her last three events. Faust is also trending highly, with recent wins over Blue and Skerzo at his last two tourneys, and a relatively favorable projected showdown with Wally before Top 64. More on that later.
I will say – CPU0 and Leffen are projected to play before the Top 32 of the winners bracket, which makes for a confusing analysis of having them both be candidates for outperforming their seed. It’s likely that my recreated bracket happened to not factor in the projected difficulty of the losers bracket. As of right now, the loser of that match is set to play Justus into Slowking, n0ne, and Spark before Top 24, which looks quite hard. Conversely, Leffen is also cracked. Even assuming he loses to CPU0, Leffen will likely beat all of them. Maybe CPU0 is just that cracked too.
The Struggle Bus
Here’s who didn’t fare nearly as well: Wally, KoDoRiN, S2J, Preeminent, Raz, JSalt, Grab (whom it gives me no pleasure to point out here), and Zanya. They overwhelmingly have very difficult brackets. In Wally’s case, Faust is a really tough draw and if history’s any indication, it’s one he may outright DQ from. KoDoRiN, S2J, Preeminent, and Raz have quite hard matchups in their projected winners brackets (Axe, Spark, Skerzo, and n0ne) before they even reach the highest seeded opponent of their projected runs (aMSa, Zain, Cody Schwab, and n0ne).
Meanwhile, JSalt, Grab, and Zanya have tough respective outs in their projected winners brackets (Zamu, Krudo, and Ossify). Even their projected losers bracket opponents, assuming they make it to Top 64, are not freebies (Dacky, Kacey, and Graves). Beating all these opponents – or whomever else takes their spots – would be impressive enough, and yet in terms of placements, it would also have them finish within their expected seeding range.
Keeping these players in mind, I created another graph somewhat similar to what I made above. The difference is this time, I used KMeans clustering to create four different clusters within my Top 64 in order to create generalizations about our player groups.
Within each of these clusters, you’ll notice a clear pattern as well. For example, Cody and Zain’s chances of doing well are far beyond the field; within my simulations, the two of them combined for more simulated victories (357 for Zain, 320 for Cody) than everyone else. Honestly, the graph above underplays the gap between them and everyone else because it merely focuses on average placement. Significantly further away from them in terms of wins is aMSa (51), Plup (48), moky (45), and Jmook (41). Beneath that, we have some interesting cases. Though he ended up with one of the highest and most reliable average placements, Hungrybox won the event only 21 times, lower than Leffen (35) and Aklo (24). Meanwhile, each of Wizzrobe (14), Trif (13), Soonsay (7), and Joshman (5), hung around the one percent miracle territory. Broadly speaking this whole group is the ‘fringe of major winning contention’ to ‘the best of the best’ tier.
Granted, this is not everyone who won a major in our simulations. Beneath this group, you have essentially reliable Top 50 players whom you can expect to see deep in the bracket, if not fighting for top eight. Within these people, lloD won five times, S2J won thrice, Salt won twice, and each of Spark, Krudo, Fiction, and Junebug won once. The model was not nearly as kind to KoDoRiN, Axe, Medz, or SDJ, likely due to the inherent lack of “winning” paths; none of those players finished in first place.
We then get into the “wild card” range of everyone from Ossify to Kacey, most of whom I addressed earlier in the column. These are basically people who are either high seeds with very tough brackets or low seeds with relatively high chances of making potential upsets. For the most part, we’re out of the major winners, though somehow, each of JSalt, mayb, and Frenzy had simulations where they took home the gold. I like each of these players, but I’ll be real: nothing crazier in Melee has happened than JSalt, mayb, or Frenzy winning a major. When you simulate a tournament 1000 times, sometimes your model spits out random shit.
Lastly, there’s the “buzzsaw of death” of players whom I dub as such because they have to play a top seed. Everyone from Kürv to E-Tie is unfortunately in a range where advancing is a tall order. It’s not impossible for a player of this type to make a huge upset – earlier this year, Kürv technically defeated Jmook at a local – but it’s extremely unlikely. Of these players, the biggest outlier performances were fifth place runs from each of OG Kid, Graves, and Dacky. In the overwhelming majority of simulations, they did not make it to Top 32.
Considerations & Limitations
As much as I would love to pretend that my process was even better than last time, in some ways it was not. In fact, there was one particular error that I noticed, which admittedly muddies the waters as far as my top and bottom seeds are concerned – a conspicuous lack of 49th place finishes for my top 16 seeds and my bottom 16 seeds. In their place are entirely 65th or 33rd place finishes; my guess is that this is because of my flawed bracket structure, with preset matches and no losers bracket adjustments to avoid ‘double jeopardy.’ This was a problem with my last simulation also.
To be fair to myself, this will take some time to adjust. Still: I think the charts above do a good job normalizing possible placements and performances at the event given our win-rates, and I think I can achieve similar levels of predictive success to my previous model (84.3 percent success rate for placement ranges, and 12.4 percent higher than if I just predicted sets by seed). Given that the primary goal of this model is to predict who the most likely candidates are to win the event, I am fine with this somewhat big, though less relevant, stipulation.
The last asterisk is the likelihood of DQs. I simply forgot to put this in. If it’s any difference, based on insider knowledge I have (kill me), I believe that Wally and E-Tie, bare minimum, are going to DQ from the tournament. That number could go up or down depending on how the rest of the week turns out – I don’t really know – just keep this in mind. Frankly, this is probably my biggest flaw in methodology; I should just always assume that a DQ will happen and use past data to run a quick ‘check’ on players before simulating a bracket (and replace them with a generic stand-in player if they fail the check).
Who is my Pick?
Heading into this column, I wanted to go with Zain. The reasoning is pretty simple: he is not likely to play aMSa, and when he is not likely to play aMSa, he typically wins the event. We’ve seen this for almost a year and a half now. It’s not rocket science. However, I think I’ve earned myself a little bit of credit. I don’t want to go with one boring pick here. I want to go with one of the two boring picks; it’s a matter of which one.
All in all, this is a great bracket for Cody. Before top 64 begins, he has two projected Fox dittos vs. Certified (or RyGuy) and salami (or Lowercase hero). These are strong players that Cody will nonetheless proceed to stomp into the ground, because that is what he does to spacie players within the Top 100, let alone outside of it. After that, he will have yet another Fox ditto against one of Preeminent and Skerzo, and you can take a guess as to what Cody will do then.
This brings us to Top 16 from the winners side, where Cody will play one of Junebug or Salt. Although it’s tempting to envision Cody losing to another Donkey Kong as a possibility, based on his last set with Junebug, I wouldn’t count on it, and it seems very unlikely that a Captain Falcon under Wizzrobe would defeat Cody as well. Guess what he has after that: a third Fox ditto (or potentially fifth straight fast-faller matchup) with moky. It’s not a sure-fire win, but we know Cody will be plenty warm for that matchup, and that’s assuming moky gets by Trif. If moky does not get by Trif, we’ll then see another opponent that Cody has, in recent memory, viciously beat down. His reward for defeating a bracket composed of fast-fallers or people he’s destroyed – aMSa in winners semifinals. Respectfully, I don’t have to explain that one.
Obviously, a Zain/Cody showdown is a coin flip, maybe even one in which Zain holds a slight edge. But this bracket totally screams “Cody win” to me, even if Zain technically has an advantage in both the head-to-head and for major victories. More than that, Zain has nothing to prove here. He has effectively sealed the deal on number one with enough decent showings for the rest of the year; the margin he has over the field is basically impenetrable. Or is it? Cody has nothing to lose here either. He’s publicly said he wants to take most of next year off. However, Cody also has everything to gain. It’s not too different from where both Cody and Zain were heading into Shine 2023, where Zain had this huge lead in majors, and Cody snatched victory from the jaws of defeat at that tournament. We saw how the story ended: a first-to-ten for No. 1 where Cody took the crown.
This situation is a bit different. Unlike last year, in which the two were tied in major wins, Zain’s won six of them in 2024, and Cody has three, with two few events (DPOTG, the LACS invitational, and the Nouns invitational) to go. It would take something even more unprecedented from Cody and a collapse from Zain for him to lose his stranglehold on the No. 1 race. But no matter what, you can never discount the probability of Cody winning a supermajor. There’s also a final X-factor to note: Mang0 is not here. At the risk of having a bunch of Internet people get mad at me again for being a Mang0 fanboy – which, I have to say, is very stupid but also very hilarious given that I’m also accused of being anti-Mang0 because of his qualms with the rankings – this undeniably will have a big impact on the competitive field. The long story short – it gives Cody an edge in terms of his chances to win a major while strictly hurting Zain’s chances too.
If there’s something I’ve learned from months of failed major predictions and successful ones alike, it’s that predicting who will win a major is quite hard. Ironically, it’s also very simple: you take note of head-to-head trends across the field, compare them among the relevant players, and roughly examine how brackets could possibly turn out. Winning a major can be predictable – simply just pick the top seed every time – but a large part of it comes down to luck. Who are your most likely opponents? Who are your toughest opponents? Who are your toughest opponents’ toughest opponents? What’s the most likely bracket of this event? Are you more likely to win a matchup after having already played it in a previous round?
The answers to these questions don’t always neatly align with our preconceived ideas of how to assess them. In fact, they are not always reflected by the numbers either (or not the ones we have right now). However, when it comes to predicting events, I tend to imagine them as a series of “edges” that a player might have or not have. In the case of DPOTG 2024, Zain is the slight statistical favorite, but sometimes there are edges you can’t quite quantify. I think Cody Schwab has enough of them to win DPOTG 2024.
Appendices: