With a quick overview of Eventum Network done in my previous post, I decided to write in detail about how their market reward mechanism functions. Those that already participated in previous events know that to earn a reward, they need to report data correctly and in time to join in a consensus, which is also known as general agreement (majority, by default, is set to x>2/3 of participants). Once in a consensus, the developer distributes the staked reward to everyone in the consensus who provided verified and real-time data. But how exactly is the reward calculated you might ask?
If the reward were to be distributed evenly, everyone would earn a reward equal to C/n , which is the total reward (C) divided by the amount of participants who formed a consensus (n). This is called an average reward for everyone in the event, but the exact amount deviates from the average according to their submission speed and other factors. Therefore, those who are on average faster than the rest receive a proportionately bigger reward than others. As more and more people join events (and later markets), the average reward will continue to decrease (if the reward stays the same, which it wont). This was the case before Eventum went live, when the highest yielding reward on the event was at 9,15%.
According to the whitepaper, the share of the reward is distributed with an exponentially decreasing function. The formula provided by the whitepaper is C/n * speed factor * market join bonus + C/(N+k) (stake that gets returned). Market join bonus is optional, depending on the time the participant joined a market. Some users have also reported a problem where they submitted the correct answer but did not form a consensus. This is because they have submitted their answers after the consensus is formed and as a result are not entitled to a reward (and would defeat the purpose of Eventum).
I will be analyzing the Dota 2 event that commenced on February 25th, because I gathered the most data out of all events. The total amount of EVT and ETH distributed in the event was set at 900.000 EVT and 6 ETH respectively. I calculated the average reward based on the formula from the whitepaper I described above at 0,65% (there were 155 people in consensus). That means on average, every participant that formed the consensus received 5850 EVT tokens and 0,039 ETH. Because the distribution is not equal, the highest earner from the data at rank 4 received 2,57% of the total reward (23130 EVT + 0,1542 ETH) while the lowest earner from the data at rank 255 received 0,1% of total reward (899 EVT + 0,006 ETH). I suspect Eventum is using a heavily modified version of Pareto distribution, also known as 80–20 rule, which states, among others, that the top 20% of population holds 80% of total wealth in a given society. This goes in line with what I’ve observed in Eventum events, which favors those that submit the correct answers early, incentivizing the real-time nature of Eventum. Let us start with linear regression first.
Judging by the first graph which shows the linear trend line, the coefficient of determination (R squared) shows that 50,67% of total variance can be predicted from the independent variable. The higher the R squared, the better the regression line will fit the data points and the better explanatory power the model will have. For the linear regression, for every increase of rank by 1, reward declines by 0,65% on average, ceteris paribus. I included it in the model because it is a textbook example of basic regression models, but it ultimately is not fit for this type of distribution. Because we deduced from the whitepaper that the distribution is an exponentially decreasing function, I plotted the following graph.
This is an exponential trend line and the regression line is much more fitting the data I have. For example, 82,8% of total variance can be explained with the independent variable and for every increase of rank by 1, the reward decreases by 0,005%, ceteris paribus. While this model can explain the data quite well on the long tail, it fails greatly in ranks up to the first 25. Next up, I tried the power distribution.
The power distribution with the provided data much more nicely. For every increase in rank by 1, the reward decreases by 0,007% and it explains 95% of total variance. I believe this is the best approximation of the function Eventum is using, but my curiosity got the better of me so I decided to test the logarithmic function as well.
This model explains over 95% of total variance, which is a tad more over the previous distribution. For every increase in rank by 1, the reward decreases by 0,01%. This model seems superior to the rest, but limited data is preventing me from doing more analyses. In other events, polynomial function best fits data (but is ultimately false), in others exponential etc…
But what’s to stop others from gambling by guessing the outcome they need to report and earning a higher reward? One factor is that developers can require users to provide data with unique properties (e.g. Who killed whom at which specific time in Dota 2?).
Another factor which prevents gaming the system is to require users to stake a fixed amount of tokens based on the level of their reputation score. Every user starts with a reputation level of 3, and a persistent low level over time flags that user, preventing him from joining events in which developers require high reputation level, or staking a proportionately higher amount of EVT tokens. If the user guesses wrong, their stake goes to the reward pool from which the users in consensus are rewarded. Staking mechanism is defined by , where C is defined as total reward, N as the amount of people that joined the data feed up to that point and K as the stake factor set by developer (which ranges from 1-5 — 1 being in a high fluid market with lots of users, and 5 in a small market with few users). Stake factor can potentially be much higher than the average reward and is there to prevent illicit activities in the system. Small reputation penalties (f = -0.5) are given to users that join but never report the data in a market, medium penalties (f = -1) are given to those that report data but do not form a consensus and are with three standard deviations from the mean and big penalties (f = -3) to those that report the data, do not form a consensus and are outside of thee standard deviations from the mean, which means that they were trying to guess a result by reporting very early. For every violation, a multiplier is added that punishes the user.
In the event that a wrong consensus is formed that is based on false information (intentional or unintentional), a percentage of data providers data did not form a majority will receive a notification that their data was incorrect — who receives this notification is based on their reputation and chance. After these users receive this notification, they have an option to open a dispute and oppose the consensus. To do this, they need to stake a large number of their tokens (based on their reputation level — to prevent abuse) that are returned to him in the eventuality the dispute is successful. In the event of a dispute, a number of data providers not present in the initial consensus are invited to form a second consensus called a dispute resolution. If the second consensus differs from the first one, the reward is taken from the first group along with their stakes and awarded to the users in the second consensus. The reputation of users in the first consensus decreases while the reputation of the second increases. To further reduce the chance of fraudulent activities, the Eventum platform employs machine learning algorithms (network statistics — latencies, node origins; behavior analysis — voting patterns, same markets, submission time) to recognize dishonest behavior and asks (and rewards) random users in similar markets to confirm suspicious activities in a given market.
According to Martin Mikeln here and here, Alpha version of their platform is already supporting the reputation level, but they are still working on staking and dispute & review mechanisms. They hope that the reputation level will already affect users in future events to prevent people joining that have already gambled in the previous events by submitting their answers early. However, I believe that guessing in the next two events will be minimal since they will have to recognize false information from Eventum whitepaper, which is quite hard to if you haven’t reviewed it beforehand. But even the implementation alone would be impressive to see. Overall, the team is doing impressive works even before the ICO, so I am quite enthusiastic about the end result they will deliver by the end of Q4 this year.
FOR MORE INFORMATION:
Website: https://eventum.network/
Alpha version (MVP): https://alpha.eventum.network/
Whitepaper: https://eventum.network/Eventum-whitepaper.pdf
Lightpaper: https://eventum.network/Eventum-lightpaper.pdf
Telegram: https://t.me/eventum_network
Twitter: https://twitter.com/eventum_network
Reddit: https://www.reddit.com/r/Eventum/
Facebook: https://www.facebook.com/EventumNetwork
Medium: https://medium.com/@EventumNetwork
Published By : bheem0
bitcointalk : https://bitcointalk.org/index.php?action=profile;u=1130931