## Triumph's Data Driven Solution to Skill Gaming Compliance

#### Triumph's Data Driven Solution to Skill Gaming Compliance

Our goal at Triumph is simple: to enable developers to effectively monetize their games through real money esports competitions. To offer a seamless, plug-and-play integration experience, our service must fastidiously shield developers from exposure to the complexities surrounding real money offerings. In short, we shoulder the burdens and take care of compliance so you don’t have to. The most formidable of these challenges also happens to be the most frequent question we are asked-- “how is this legal?” The Legality of Skill Based Wagering In short, skill based gaming is not typically gambling. The long answer is that Triumph takes a multifaceted approach to assuring legality in as many U.S states as possible. We partner with best in class legal counsel, automate know your customer (KYC) processes, do meticulous bookkeeping, and utilize data driven skill assessment tests. Before delving into Triumph’s approach, we need to understand the skill gaming landscape at large. Starting from first principles, two fundamental questions arise: (i) what is a game of skill, and (ii) why are skill based games legal? To answer the latter, gambling is defined at the state level, and broadly must have all three of the following elements: consideration (a buy-in), prize (a pay out), and — the most important defining characteristic —chance. Skill based gaming allows the first two elements while disallowing chance. If you’ve ever entered a golf tournament, played a carnival game, or ran a competitive marathon, you’ve participated in a skill based tournament. Precisely defining chance and skill, however, is a bit more difficult and far more interesting. The former requires a more in-depth explanation. What is a "Game of Skill"? Skill is a nebulous concept that evokes a visceral feeling that is hard to rigorously pin down. Because gaming law is adopted at the state level, there are various interpretations of the word “chance,” which may make it seem difficult to discern what is and is not permissible. To add to this complexity, most modern games have at least some interplay between skill and luck. While chess’s perfect information and lack of randomness makes it intuitively “skill based,” analyzing a more complicated game such as Fortnite is not as simple. Sure, what you get in the next treasure chest may be random, yet it is clear that I would not be able to land one shot on world champion Bugha in a thousand games. All of this becomes even more murky when considering the plethora of bad actors in the skill gaming space, who attempt to feign luck in “skill based contests” by using dubious design elements including: dice, cards, coins, casino-like color schemes and sounds, and other gimmicks. All of this is to say that it is impractical for us to qualitatively analyze each and every source of randomness in a game. Because of this, we've decided to abstract away actual gameplay and classify games only using data driven techniques. This proves to be far more foolproof and definitive than other methods we've seen. Triumph's Approach We first look at the qualitative elements of a game and ascertain all elements of potential randomness and information asymmetries. If we can remove these -- we often can by synchronizing in-game random number generators, as explained in our documentation -- we are done. In most games we go live with, we are in fact able to do this. In cases where we cannot, we turn to a dataset of all in-game matches. At this point, we utilize multiple skill tests to get a holistic picture of the degree of skill a game requires. Some of the tests we use include: - Utlizing an Elo based system (often used in games like chess) and measuring the variance of Elo population wide. Since differences in Elo imply a theoretical win percentage -- for example, if my Elo is +200 over yours, I will win 76% of games in expectation. The larger the variance in Elo, the more people we can "chain" together who have high expected win probabilities over the next person, suggesting a game is more skillful. As a counterexample, a coin flip tournament would have an expected player Elo variance of 0, given enough games. - Measuring player improvement over time. Do player's scores increase over time, as they learn the intricacies of the game? What is the rate of increase? Is it predictable? Note: this has the added benefit of helping us with fraud detection. - Using network/graph theory results to make a flow graph and measure the proportion of "cycles" in a graph, with each player representing a node, and each edge between players representing a net win/loss value. The higher proportion of cycles, the less evidence there is that the game is skill based. For example: a game where player A beats player B, player B beats player C, and player B beats player A, is less indicative of skill (note: this will happen sometimes in practice even in a highly skilled game) than a game where player A beats player B and player C, and player B also beats player C. - Perturbing observed game results with some forced "noise randomness" and seeing if the functional distribution from which our perturbed results come differ drastically than the real results. The more that noise impacts our Elo calculations, the more signal we have that the original results come from a game of skill. For example, in a coin flip game (with no skill), adding forced randomness by switching real game results with a random winner will have no measurable impact, since the results are random to begin with. Alternatively, switching the results of my game with Bugha in the aformentioned Fortnite example with significantly alter our calculations. - Measuring player score variance relative to population wide score variance. In a skill game, we'd expect the population wide variance to significantly exceed the average of individual players' score variance. - Measuring outliers and watching the distribution of scores. Depending on the game, we use some or all of these tests to try to grasp the skillfulness of the game. Using a multi faceted approach is in our view the best way to pin down such an ill-defined yet intuitive concept. What Else? This quantitative test gives Triumph unquestionable advantages. Most importantly, we can be certain that we (and you) are operating on a sound legal basis where we operate. Secondly, our test is scalable -- it works with any game independent of its specific mechanics -- we simply need players’ scores in head to head matchups. It actually gets more accurate as the number of partner games we include increases. Beyond this, we (i) run strict KYC and geolocation, ensuring that users’ are 18+ and playing from a state where skill based gaming is permissible, (ii) have a long reasoned legal opinion written by Duane Morris, that reviews our offerings on a state by state basis, (iii) have in house legal counsel, and (iv) employ industry leading skill matching. Long story short, you can sleep comfortably at night, knowing that Triumph has taken reasonable measures to reduce your risk. Jared Geller Founder and CO-CEO