The Wild West of E-sports Stats

Late last year, Madison Square Garden hosted WWE Live, which the self-professed global leader in sports entertainment, with its customary restraint, billed as an "earth-shattering special event." With close to 20,000 fans in the stands and many more streaming on the WWE Network, the Beast Incarnate, Brock Lesnar, beat the Big Show in Lesnar’s first MSG match in more than a decade.

On the same October day, in the smaller, 5,600-seat Theater at Madison Square Garden, the Electronic Sports League kicked off ESL One New York, a two-day tournament featuring eight of the top teams in Dota 2, which along with League of Legends forms the fast-growing backbone of the multiplayer online battle arena (MOBA) genre, the most popular (and most lucrative) e-sport. The Dota matches weren’t scripted, and the combatants were younger, smaller, and less spandexed than WWE’s superstars. But ESL’s spectacle, which played to a packed auditorium and streamed to Twitch.tv spectators, relied on a similarly larger-than-life blend of real and pretend, featuring its own semi-supernatural heroes fighting for supremacy in a square arena.

The ESL stage was bathed in the electric blue that was beaming from spotlights and painting the first few rows of the audience the color of Cerenkov radiation. Most of the hallmarks of traditional sports were on display. A panel of besuited and beheadsetted studio analysts, some of them former pro players, spouted sport-specific jargon that would have been completely impenetrable to anyone new to the game. An on-camera emcee solicited clichéd comments from Peter "ppd" Dager, the 24-year-old captain of Evil Geniuses, the team that had won the fifth edition of Dota 2’s annual championship, The International (TI), in August and taken home more than $6.5 million in crowd-funded winnings. (The pivotal play was an earth-shaking special event. Suck it, Vince McMahon.) And during breaks in the action, corporate sponsors screened targeted ads for gaming mice, video cards, pizza rolls, and Soylent, a liquid meal substitute now marketed to gamers for whom pizza rolls require too much preparation.

For all the commonalities, though, ESL One still differed from the Garden’s regular draws. Some of the five-man lineups had turned over almost entirely since TI thanks to incestuous swaps: Four-fifths of the talented-but-combustible Team Secret had walked away, leaving the lone holdover to rebuild. Even though it had won a TI title, Evil Geniuses had rebuilt, snapping up one of Team Secret’s departed, Arteezy, to replace the popular Aui_2000. (Later, the two would trade places again.) Another team, Archon, had built its starting five by absorbing an existing squad a day before the tournament. "We’re very happy to welcome Jeyo, Fluff, 747, Whitebeard, and Moo," the press release said.

The biggest difference, though, surfaced amid the studio hosts’ standard old-sport narratives about big stages, must-win games, and grudge matches. Over and over, the analysts returned to one X factor/question mark: 6.85, the latest update to Dota 2, which had dropped a week earlier. Dota updates adjust items, mechanics, and character skill sets in order to maintain the balance that players constantly seek to disrupt. This cat-and-mouse metagame spawns an endless cycle of heroes being buffed (made stronger) or nerfed (made weaker) as public and pro matches reveal attributes that are under- or overpowered. Imagine a universe where the NBA could not only nerf Steph Curry’s long-distance accuracy rating or LeBron James’s strength and awareness scores from season to season, but where it had precisely constructed those players in the first place.

Although 6.85 was considered a conservative update that wouldn’t dramatically destabilize the way the game was played, the patch notes ran almost 3,000 words, and many of the tweaks had implications that weren’t completely clear — even to the game developers, who can’t always anticipate the sometimes-serious consequences of MOBA "fixes" once they’re in the wild. "It’s anyone’s game depending on how they’ve adjusted to the new version," one announcer said, setting up the first match.

The eventual winner at ESL One wasn’t Evil Geniuses or Team Secret, but an unheralded entrant called Vega Squadron. According to pre-tournament Monte Carlo simulations based on Dota 2 analyst and broadcaster Ben Steenhuisen’s Elo ratings, Vega — a team that had stuck with the same lineup that had failed to qualify for TI — had a 0.2 percent chance to be the last team standing, compared to favored EG’s 58.1 percent. But those probabilities were based, by necessity, on outdated information. By the time ESL One started, less than three months after TI, neither Vega’s opponents nor the game itself was the same. And as The International 6’s main event begins today in Seattle, only three of the eight ESL One squads have qualified for the 16-team bracket — and those three combined have replaced five of the 15 players who participated in New York.

That constant mutation and team turnover is part of what makes MOBAs unpredictable and thus compelling, but it also creates complications. In the last decade, rapid growth in audiences’ understanding of statistics and insiders’ appetite for applying them has transformed how the world’s major sports are played, broadcast, and consumed. The financial incentives for electronic competitors (and the infrastructure surrounding them) are coming to mirror those of more mainstream sports, which would suggest that it’s time for e-sports stats to take their turn in the spotlight. But a genre whose young, computer-friendly fan base makes it seem uniquely well suited to statistical analysis is in practice difficult to crack, despite its ability to borrow breakthroughs from other sabermetric movements. And as the flood of information that digital games deliver arrives earlier in the industry’s life cycle than big data did in longer-established sports, e-sports analysis is struggling to keep pace.

We’re past the point of marveling when an e-sports event sells out a stadium known for housing more established sports, and we’re mostly over the semantic debates about whether e-sports are sport-sports. The U.S. government says pro gamers are athletes, and perhaps more important, media companies are starting to treat them accordingly: In a little more than a year, ESPN went from making disparaging checkers comparisons to broadcasting tournaments, doing a documentary about TI5, and launching a hub for e-sports coverage. But even in 2016, when e-sports are, by many measures, mainstream, MOBAs demand explanation in a way that other MSG staples don’t. That’s partially because e-sports haven’t permeated American culture to the degree that they have elsewhere in the world — although they’re more popular than baseball among male teens in the U.S., according to an ESPN Research and Analytics study — but it’s also because they’re deceptively complicated, with too many variables to summarize in simple loglines like "tall people try to throw a ball through a hoop."

The diagram below shows the layout of a typical MOBA map:

The ultimate goal in any MOBA match is to destroy the opponent’s base, but doing so isn’t as simple as having quicker reactions or better aim. Each five-player team tries to gather resources by "farming" (killing weak, computer-controlled characters) and destroying human opponents as quickly and efficiently as it can in order to acquire gold and experience, which don’t directly lead to victory but can be redeemed for items and abilities that eventually give one side the superiority to mount a successful base assault. Travel between bases takes place largely in the three "lanes," which are guarded by defensive emplacements and AI "minions" or "creeps." Each lane caters to a particular character type or play style, based on map layout, objective locations, and proximity to other players.

Teams fill out their lineups during the drafting phase, the game before the game in which each side selects from more than 100 heroes (or in League lingo, champions) while the other side seeks to anticipate (and ban) the heroes it doesn’t want its opponent to use, like lawyers selecting juries. Each unit tries to maximize its players’ strengths by paying close attention to team composition, crafting a balanced squad or a roster that’s built to excel early or late. At the pro level, victory requires intense training and attention to detail, tactical thinking, and constant communication.

MOBAs are big business, with hundreds of millions of amateur players, many of whom idolize, envy, or aspire to join a small group of experts who play in high-profile pro circuits and compete for endorsement deals and seven-figure follower counts and prize purses. (If it strikes you as strange that so many people would want to watch other people play video games, think again.) The bigger the payouts, the more powerful the incentives to gain an edge on one’s opponents. "I’m pretty sure that some of the Dota players who won [last] year’s International have earned more money from that than a professional StarCraft player earned from his prize money for a lifetime of playing StarCraft," says Steenhuisen, referring to the best-selling real-time strategy game that helped birth MOBAs and popularize e-sports. "So we’re growing at such a rapid rate that we’re reaching the point now where an [in-house] analyst that can affect the team’s chance to win against any other team by 1 percent, it’s converting in tens of thousands of dollars. It pays for itself."

Appraising that percentage is still an imprecise science, but the rising stakes suggest we’ll soon see e-sports follow the familiar, multistage analytics epiphany that’s played out in other sports. First, there’s the homebrew phase, in which loosely organized hobbyists, some of them pseudonymous, workshop their ideas at personal blogs or message boards or bean factories, either ignored or outright ridiculed by their sport’s risk-averse establishment. Eventually, someone on the inside sees the value in the analytical community’s counterintuitive conclusions, profits from picking the low-hanging fruit, and welcomes the outsiders into the fold, starting a run on internet analysts of the sort we saw during the NHL’s "Summer of Analytics."

Even as the genre mounts its assault on American mind share, though, the MOBAmetrics movement is still in the farming stage, grappling with technological and analytical challenges that bear only a passing resemblance to those in other sports that are further along in their life cycles. "In terms of the understanding of statistical metrics, we are at about the place that baseball was right as Bill James’s work was starting to be accepted," says Alan "Nahaz" Bester, a white-haired associate economics professor at Western University who’s carved out a side career as a Dota analyst and commentator.

The funding, at least, is flooding in, courtesy of consonant-skewed startups and wealthy ball-sport stars. Last June, billionaire entrepreneur and Dallas Mavericks owner Mark Cuban invested in Unikrn, an e-sports betting service, and made his public League debut in November. "There is definitely a place for analytics [in e-sports]," Cuban says via email. "The challenge is that every game is different. Not just the game itself but how the IP is handled."

Cuban, whose Mavs led the NBA bracket in ESPN’s "Great Analytics Rankings" last year, became the earliest American adopter of — and, later an investor in — an Australian company called Catapult, which uses GPS devices to track human movement in hopes of keeping athletes functioning at full efficiency. Every major American sport is exploring or already employing some means of player- and/or ball-tracking technology, which has opened up new avenues of research and led to discoveries that weren’t detectable with less granular data. In theory, in an increasingly quantified world, e-sports have an advantage over nondigital enterprises: They’re the one kind of competition in which player movement matters but player tracking doesn’t raise privacy concerns or require either wearable devices or a complex camera-radar system that may struggle to record certain events.

"When you think about baseball or basketball, they’re actually having to create new technologies in order to be able to create this data," says Chris Hopper, who leads Riot’s e-sports analytics team. "So they have to create a geo-tracking service that could say, ‘OK, in this case he made the cut-off throw correctly, or he didn’t.’ It has to be a smart enough program to analyze that. That’s actually a very expensive program to develop. From our side, everything’s already digital."

Hopper’s team has records of every match played since the start of the Riot-sponsored League of Legends Championship Series in 2012, although the company’s data collection has widened and deepened in the last year to preserve player locations and actions on a moment-by-moment basis. He hopes the low barrier to big data will shorten the time it takes for e-sports analysts to catch up with their intellectual cousins in more physical sports. "Baseball has been around for a hundred years, but it’s only really been in the last 15 or 20 years that you’ve seen just a glut of new statistics come out, which tell such a greater story of the game," Hopper says. "Our goal is to touch on that level of sophistication without necessarily needing that kind of lengthy timeline."

The potential benefits to teams are obvious, but Hopper believes the most important application of e-sports data analysis is helping spectators "understand what they’re about to see on the screen," which deepens fan engagement with the game. "In a lot of cases, you could look at our players and at a very high level, their numbers will look very similar," Hopper says. "OK, maybe this guy has more kills, maybe this guy tends to have more assists than the other guy, so his team fights more. But we can dig deeper, and we can find these little grains of wonder that will really kind of astonish the [spectators] and lead them to think, ‘Oh man, that’s a cool thing.’ And then they wonder if that will happen, so then they’re looking forward to something."

There are, of course, a couple of catches. The first is that the samples are on the small side: As Hopper explains, League’s championship series is split into two 90-game seasons, plus playoffs, which brings the total number of games to roughly 220 per year. The other is that much of the data Hopper has isn’t accessible to outside analysts. Instead, most make do with end-game statistics — total kills, total deaths, total assists, and so on — which Hopper likens to baseball’s box scores. Hopper says Riot’s restrictions stem less from a desire to keep the data private than from making a smaller early investment in the game’s information interface than Valve (Dota’s developer) did.

Valve’s relative transparency has made it possible for Dota statisticians to delve deeper and draw more useful conclusions. If League allows users to see the equivalent of MOBA box scores, Dota provides the equivalent of MOBA play-by-play, via replay files that can be parsed to reveal the state of the match at frequent in-game intervals. "Where we are right now in Dota is we’re making the transition from just looking at end-game statistics — kills, deaths, assists, gold per minute, experience per minute — to looking at statistics as they evolve throughout the match," Bester says.

Using this time-stamped information, sites such as Dotabuff can generate displays like this one, which shows the minute-by-minute gap in gold between Team Secret (red) and the victorious Vega Squadron (green) in the last third of their decisive match at ESL One New York. The Gordon Gekkoesque takeaway: Gold is good.

Dotabuff also hosts interactive maps that combine temporal and spatial information. This map of kills from the same match, which Bester compares to an NBA shot chart or an MLB spray chart, shows how concentrated both teams’ deaths were on the darker, "Dire" side of the map, where Team Secret’s base was — an indication of how effectively Vega pressed the attack.

It’s tempting to reduce Dota and League to virtual versions of other sports, which can be dangerous. "It’s not obvious that certain things are transferable, and in some cases things that would seem obviously transferable are not," Steenhuisen warns. But even the people who urge others to stay away from sports analogies can’t seem to stick to their own advice. In the span of a single exchange, Cuban mentions that MOBAs are "more complicated and not analogous to ball sports" but then makes a ball-sport analogy himself, explaining that "for League of Legends trying to do analytics is on par with trying to do analytics of the quality of quarterback reads."

He’s right. MOBA analysis suffers from the same quality that hampers analysis in almost every sport but baseball: With 10 heroes/champions in constant motion on the map, it’s tough to isolate the contributions of any one player. "A mid-laner might be winning mid-lane but end up having a terrible stat line at the end of the game because his top laner got smoked and the enemy top laner killed him a bunch of times," Hopper says. "All five positions matter, all five players are playing at the same time, and that complexity, we feel, is something that can’t ever be totally reduced to a numerical basis."

The genre’s existing stats come with caveats and causality concerns. Win rates and "relative to average" stats don’t account for team quality or the quality of opponents, which vary much more in MOBAs than they do in major U.S. sports. KDA (kill/death/assist) stats don’t factor in how often a player stays hidden from an adversary’s view, which forces opponents to play more conservatively. And the community continues to debate how positions should be played. Evil Geniuses won TI5 with two players, Aui_2000 and the versatile, Ben Zobristesque Fear, who approach support play from opposite poles on the selflessness spectrum: Aui_2000 puts his heroes at risk to go for gold, giving him high death counts and an unusually high "farm share" for a support player, while Fear (who played as a "core" hero at TI) focuses on protecting other players while they get the gold. "We can quantify all these things, but we don’t know if it actually correlates to success or not," says Bryan Herren, a statistician for e-sports broadcaster Beyond the Summit. "Dota is the Wild, Wild West when it comes to this kind of stuff."

Appropriating other sports’ stats is one way to impose order. Steenhuisen is experimenting with neural networks, which have helped in baseball and basketball. Bester, too, believes he’s coming closer to distilling the disarray into a true value metric by drafting in the wake of earlier work. If you’ll allow another lapse into ball-sports analogies, the MOBA Baseball-Reference is Datdota, another one-programmer compendium of every imaginable stat. By parsing thousands of replay files, Datdota’s founder, Martin Decoud (who recently transferred the site to Bester and Steenhuisen), gave analysts the tools to turn the descriptive into the predictive. The information it offers allows for midgame win predictions that outperform prematch, team-level Elo ratings, which lose some of their value whenever a roster’s composition changes. Theoretically, it also gives statisticians a way to credit players for portions of a win by studying how much each of their kills contributes to a gold/experience advantage, which in turn improves the odds of a victory in a quantifiable way.

Analysts of other sports that aren’t blessed with baseball’s stop-motion sequence of discrete events have coped with the clutter of continuous action by building "With or Without You" (WOWY) models that evaluate players based on team performance while they’re in (and out of) the game. This is harder in MOBAs than in basketball, because basketball’s objectives aren’t distinct from its points system, and because MOBA teams don’t make midgame substitutions. But Bester plans to synthesize baseball’s win expectancy and basketball’s plus-minus to produce a player valuation model for Dota 2. "Once you have this team fight data, you can actually track which hero has participated in which kills and in which engagements," Bester says. "And you can look at the difference in the average gold and XP swing for your team in the fights in which a certain player participated, and subtract the average gold and XP swing in the fights that they didn’t participate in, and get a form of adjusted plus-minus." Bester hopes that an accurate rating system will give players a better chance of being paid what they’re worth in an industry with no representative union and little infrastructure surrounding player representation. It could also act as an objective counter to any bias that helps perpetuate the competitive scene’s pronounced (albeit improving) skew toward male players.

Hopper is less sanguine about the possibility of a wins-above-replacement-style stat. When asked what the "Moneyball" of MOBAs might be — some unsung skill or strategy that matters more than the market thinks — he grimaces. "I had one of the analysts on my team look at it," Hopper says. "And he did a lot of work and talked with a bunch of pros, and came back with a list of 40 unique influencers per position. And I was like, ‘I need on-base percentage.’ The amount of discussion we had and the amount of, ‘Well, you can’t leave that out from the equation,’ indicated the complexity of the sport and indicated that there isn’t a great way to really home in on a single number that makes me say, ‘Yes, this guy is better than this guy.’"

The "too many players" problem is one that other sports’ analysts are still struggling to solve, but at least e-sports statheads can repurpose existing techniques that dispel some of the fog. There’s no such assist from other sports in lifting the MOBAmetrician’s biggest burden. "One HUGE difference and I can’t emphasize this enough," Cuban says. "Riot changes game properties every few weeks. This is a huge negative for analytics. It increases the complexity significantly." Of course, it also increases profits: Thanks to Riot’s constant character creation, the free-to-play League of Legends totaled $1.6 billion in 2015 revenue, easily outstripping its competitors in last year’s digital downloads market.

Dota doesn’t update as often as League: 6.85 was its first full patch since last April. (It’s now up to 6.88.) Still, it’s easy to see why patching poses such a problem. Even if the data that’s currently compiled on player performance and team tendencies were comprehensive, each update would reduce its predictive power. Imagine what it would mean for NBA analytics if Adam Silver could decide the Warriors were too good and tinker with their on-court chemistry, or what it would do to baseball’s pristine statistical record if Rob Manfred could drag a slider on an unseen menu slightly to the left to slow down Mike Trout’s bat speed or Aroldis Chapman’s four-seamer — or, as Herren says, decide, "OK, now we have fourth base." Those tweaks might make games more fair or entertaining, but they’d also compromise projections based on pre-patch data and limit any new analysis to a tiny, post-patch sample. For high-level MOBA players, the process is complicated even further by the fact that trends among public players may not translate to the pro tier, which makes it iffy to extrapolate from amateur usage/win rates. "Some of the things that happen in one game can completely affect how the next six months of Dota is going to occur," Steenhuisen says.

"When patch notes come out, you always read them, and you start trying stuff," says Luis "Deilor" Sevilla, League former coach for Fnatic, which maintains competitive teams in several titles. "You try everything that could work. And also you have to be watching all the other regions to see what other regions are doing, to see if you see something interesting, and you adapt it to how you play. So you’re constantly working on reinventing yourself."

Despite the patch problem, the lack of granular League data, and the difficulty of discerning who did what, there’s still a growing hunger for statistical insights in e-sports, both from fans and from professionals. "It’s built into the games and it’s built into the culture," Bester says. "I think relative to traditional sports, you’re going see a much larger percentage of the material that is come up with is going to be available to everybody, simply because I think there’s demand for it."

Since technophobes don’t gravitate toward online games, e-sports fans and competitors are, by definition, highly computer-literate: There’s no network of former stars and cranky writers who got into e-sports before it became corrupted by computers that give computer numbers. In baseball, hot-takers still accuse sabermetrics of sapping some of the sport’s beauty or mystery, reducing the divine to soulless lines of code. Code makes MOBAs exist. And the community’s players and coaches are almost universally young, which means most of them aren’t entrenched enough for their attitudes about the proper way to play to have hardened with age. "I’m probably one of the older people, in terms of how long I’ve been doing e-sports statistics," says Herren, who’s 25. "And when I say one of the older people, I’ve been doing it for three years."

Unsurprisingly, statisticians are already disappearing into design teams and infiltrating the tournament scene. To help his Dota 2 team prep for last year’s TI, Kyle "Beef" Bautista, the general manager of compLexity Gaming, hired Trent MacKenzie, an analyst associated with the Standard Deviants, a loose affiliation of e-sports enthusiasts (including Steenhuisen, Herren, and Bester) who "stat" for broadcasts and teams. "The biggest impetus was that one $19 million tournament," Bautista says, referring to TI5’s $18.43 million pool. "You want to be able to do whatever it is that you can. Realistically we didn’t even expect to get a huge edge. We just figured everybody else is going to be trying to get an edge, so we need to be on at least the same plane."

MacKenzie spent days writing stats-based scouting reports on potential opponents, pairing the numbers with videos that drove home his data. "Honestly, I was terrified," says MacKenzie, who hadn’t worked directly with pro players before but knew they don’t always take kindly to criticism. "I thought I was going to show up, send them some stuff, and just get shoved off the next day."

Instead, MacKenzie made his mark during the drafting phase. He discovered that in certain matchups, a team called Cloud Nine had a tendency to pick Io, a support-oriented ball of light that looks like the Beta XII-A entity from Star Trek: The Original Series. "Although it’s considered a top-tier hero, it doesn’t necessarily get picked or banned in a lot of games," Mackenzie says. "So if you don’t see it coming, suddenly it gets picked, and then maybe you get cheesed out in Game 1." Watching from his home in Halifax, MacKenzie willed compLexity to heed his advice. "I was sitting there, and the first bans went by and they didn’t ban Io. And I was like, ‘Aw, well damn it.’ I really thought that was important. I thought they would. And then [compLexity] picked Io, and I was like, ‘Oh!’ And they wiped the board with him. And the next game C9 was forced to ban Io."

"We all agreed that that single choice to make that pick instead of letting them have it was what won us that game," Bautista says. "We completely destroyed their strategy, and it’s not something that most people would have been expecting if Trent had not been able to reach that conclusion." Bautista says outsourcing stats to MacKenzie gave his players more time to practice and scrim, made them more confident, and possibly psyched out less-prepared opponents. "They’re walking in there with binders that have 200 pages of notes and information that they can access at any time they would want to," Bautista says. "It’s a little bit intimidating."

Even as MacKenzie’s experience proved the potential of applied MOBA analytics, another incident at TI5, which MacKenzie calls "Statsgate," exposed the liminal space e-sports statistics still occupy. In his role as TI’s official stats person, Steenhuisen, along with another analyst, compiled a 60-page "stats bible" for Valve with summaries of player and team tendencies, which were supposed to be used by broadcasters only.

In this case, an unknown caster leaked the document to one team, so in order to avoid an unfair advantage, Valve distributed the document to every team, even translating it into Chinese. (It also made its way to the press.) Naturally, teams that had devoted more resources to preparation felt that the leak had compromised their competitive advantage. Statsgate underscored the insights that stats can offer, but it also revealed that the field is far from mature. Nothing the most statistically astute announcer could say on a baseball broadcast would be news to MLB teams, but the gap between external and internal knowledge in e-sports is still small enough to cross the public-private line.

Statsgate gave any teams that were on the fence about the necessity of hiring their own analyst another reason not to. But well-financed teams might soon see analysts’ value as recruiting and talent-retention tools. Patrik Sättermon, chief gaming officer for Fnatic, notes that stability has proved elusive for many competitive teams. "You can be relevant for six months, but then the team [gets relegated] or they leave for another organization and you just get vaporized, you cease to exist in that landscape, that game," Sättermon says. "So, one of the instruments to provide stability for us is to have a supporting structure, have experts in each and every title."

Last year, Riot recognized head coaches and began subsidizing teams’ spending on coaching staffs. Fnatic has hired sports psychologists to help its adolescent players cope with the Ender’s Game–esque isolation and anxiety that come with the territory; some teams have hired masseurs to assuage the physical strain of sitting and clicking, which can cause career-ending repetitive-stress injuries. Because most coaches don’t have statistical backgrounds, analysts are also part of the ideal support structure. As a former professional poker player and coach, Sevilla is an exception — for poker players, he says, "stats [are] everything" — but Fnatic recently hired two analysts to help him pinpoint weaknesses in opponents as well as his own team. When players do depart, analysts can identify complementary replacements, although it’s difficult to forecast chemistry and communication.

"As [e-sports] gets more competitive, as it gets more popular, and as it gets more lucrative, every team will have a statistician or a full-time analyst/coach," Steenhuisen says. This year, compLexity hired Bester as a coach for TI. The team also tried to hire Steenhuisen as an analyst, but he opted to work on the public broadcast instead, citing better long-term prospects. "There’ll be loads of Dota events for years," he says. "There might not be compLexity."

It’s not easy to say where e-sports stats are headed, given that the MOBA market itself is still in its infancy. The existing titans can’t count on their supremacy: Although there are no plans for a League of Legends 2 or a Dota 3, the top two face frequent challenges from alternatives such as Blizzard’s non-MOBA MOBA, Heroes of the Storm, and its MOBA-shooter hybrid, Overwatch. What’s more, MOBAs are only one segment of an e-sports industry that also includes card games like Hearthstone: Heroes of Warcraft and shooters like Counter-Strike: Global Offensive, another Valve title that got TBS airtime (in a lousy time slot) in 2016. "I think shooter stats are going to be huge," Bester says. "The rounds are shorter; it’s much cleaner. In some sense, it’s much easier to document the guns that people are using, where they’re getting their kills, how they’re getting their kills."

In some ways, e-sports are unlike any other kind of competition that’s come before. In other areas, they’re almost identical, going through the same growing pains that force all fledgling leagues to learn hard lessons about gambling, match-fixing, cheating, hacking, drug-testing, unwritten rules, tampering, region restrictions, sponsorship bubbles, harassment, corruption, and conflicts of interest. How central a role stats play in e-sports’ future depends in part on demographics, which will evolve not only in terms of sex and national origin, but also age. As teams get smarter about monitoring (and policing) their players’ practice time, e-sports stars won’t burn out as early, which could extend the heretofore typical "late teens to early 20s" window for competitive careers. "In maybe five or 10 years’ time, you’re going to see a bunch of professionals that are 35 or even 40," Sättermon says. Those wily veterans might embrace stats to compensate for slower reaction times and stay ahead of raw rookies, whom Sevilla says often have "no knowledge of math or statistics."

The potentially lucrative market for MOBAmetrics has spawned startups like eSportsFanz and Dojo Madness, which are using data to recommend optimized builds, guide fantasy selections, and beat betting lines. But the movement’s main goals are still a few eurekas away. "We’re dealing with a problem where there are so many variables," Bester says. "We’re never going to identify them all. We’re going to have a model, because having a model is better than not, but … we’re not ever going to expect to perfectly understand the game."

Hopper believes that no matter how sophisticated e-sports analysis gets, League’s core (like Dota’s) is complex enough that it’s "not a solvable game." As evidence, he cites the fact that while MOBA bots are rampant, no one has been able to build bots that are actually good at the game.

"I think that’s what pulls everybody in," Herren says. "It’s an entity of its own. It’s a living, breathing, changing, adjusting behemoth of a competition. And no one really knows what’s the best way to play."