Return to site

What is Reputation?

by Tony Sakich, Jeremy Gardner & Joey Krug
This blog post was created to provide various ways explaining how Reputation Tokens work within Augur. We have decided to create a short video as well as an infographic to further elaborate the essential role that Reputation Tokens provide within our decentralized prediction market platform. As this is a rare blog post that features multiple authors, we are breaking the post up into three sections to ensure clarity.
We also wanted to remind everyone to check out the Augur Alpha at http://demo.augur.net or to install the full client, check the instructions here.  If you're looking for information about our crowdsale checkout here.
Section 1: Reputation 101 Video
Section 2: Reputation Token Infographic
broken image
Section 3: Everything You Need To Know About Reputation

I (Jeremy Gardner) love working on Augur. In my everything-but-programming role, I get to fly, drive, and hike around the globe and bring awareness to our platform at conferences, meetings, and lectures. The unending delight expressed by newcomers to our undertaking never ceases to invigorate me. That being said, it is always with a degree of dread that I anticipate discussing the cryptographic token used in our system, known as REP, or Reputation. My apprehension is not due to any perceived shortcomings in our software (in fact, REP is an extraordinary game theoretical coup); however, just about every crypto token (including bitcoin) is met with unrelenting, mostly well-deserved scrutiny and the last thing I want to do is make Augur come off as another altcoin/pump n’ dump scam.

For Augur, the explanatory challenge is multi-faceted. Firstly, REP is not a cryptocurrency, per se, but really much more like a score attached to some users’ public and private addresses. Importantly, REP is not an investment, it is a software tool necessary to make Augur work. Also, the vast majority of Augur users will never hold, use, or even need to know about REP. That being said, in order to have a fundamental understanding of how our software works, Reputation cannot be neglected.

So, what is Reputation (REP)?

REP is the first decentralized oracle solution ever implemented. It allows its holders to report on the outcome of events in Augur’s prediction markets, and rewards them for honesty, and regular reporting, with half of all trading fees on the platform.

Wait, what?

One of the most important features of decentralized and blockchain-based technologies is that there exists no single point of failure in a system. There is no server that can be shut down, no head that can be chopped off. However, until Truthcoin was introduced, nobody had found a way to have a decentralized system with an equally uncensorable oracle. An oracle is any mechanism that provides answers, such as a third-party API for sports games or news, or Siri on iPhones, when she decides to be useful.

Soooo, REP is an oracle solution… how?

Those who choose to purchase REP in the upcoming software sale will be required, every eight weeks, to participate in a reporting period, and during that period, spend what should take no more than an hour, to report on a random selection of prediction markets that have expired in the past eight week period. This allows Augur’s prediction markets to be closed and resolved in a fully distributed manner.

This sounds like a system prone to manipulation…

Just the opposite, in fact.

Depending on the success of our software sale (with success determined by number of REP buyers — NOT amount raised), dozens, if not hundreds, of REP holders will report on the outcome of each event. This, in addition to the randomization of markets being reported on, ensures that events are reported as honestly as possible (hence the name “Truthcoin”).  Additionally, a simple cryptographic commit and reveal prevents people from credibly revealing their reports before consensus, although we'll have to come up with a different solution once zero knowledge proofs become easy to compute.   

This is also why a crowdsale/auction is so important because it incentivizes people who buy rep to report honestly because they have something valuable at stake if they don't.  If we simply gave it away people could easily Sybil attack the system and wouldn't have as large an incentive to report correctly. 

So is Reputation a cryptocurrency like bitcoin?

Reputation can be thought of as a sort of “score” attached to an individual’s public and private address and is both divisible and exchangeable like bitcoin. However, that is where the similarities to cryptocurrency end.

Can REP be mined?

REP is not mined. It is fully distributed at the end of Augur’s software sale about two weeks after Ethereum's Frontier launch, based on the amount of responsibility (and money) individuals decide to dedicate to the software. We emphasize responsibility, because that is what holding REP requires.

What are the reporting requirements for Reputation holders?

Those who hold REP are expected, every eight weeks, to report on the outcome of a random selection of closed events/predictions in the system. This is done by simply selecting three options: Yes (event occurred), No (it did not), or Undeterminable /Unethical (which, if this option achieves a consensus, pushes the reporting back to the next period, before eventually being closed without a .5 resolution.)

How long will it take Reputation holders to report on events each period?

We expect the process to initially be quite quick, but may take as much as an hour as Augur grows in popularity.  There are mechanisms to split reporting requirements up, like requiring users to only report on a random subset of events, or branching into separate subject categories.  These will become concerns as the system scales.

Do REP holders have to research every event that they have to report on?

Not at all!

Reporting should be a fairly intuitive process, as most markets will have had time before the voting period to auto-resolve themselves (as rational actors will be incentivized to sell their losing shares before a market reaches zero value, leading most markets to have 99-to-1 odds at closing time). If markets have not resolved themselves, this may require Reputation holders to do some Googling; but more likely it is that the decision in question will have been poorly worded, undeterminable, or unethical, and thus reporters should mark the decision as such.

What happens if REP holders lie or forgot to report?

If Reputation holders fail to report on the outcome of events assigned to them during the reporting period, or report dishonestly, principal component analysis, PCA (explained at length in our blog), redistributes the lazy or dishonest Reputation holders’ REP to those who have reported both regularly and honestly. Also, REP holders will collect the trading fees from each voting cycle in proportion to how much REP they have. This makes Augur one of the most meritocratic systems in the blockchain space.

Will reporters make more money investing in REP and just selling it later, or using it to report?

REP is not an investment. If it isn’t used to maintain Augur markets, it’s redistributed to those that do use it.

Can I make a lot of money reporting?

While not the technical purpose of REP… Yes.
What happens if there start to be thousands of events to report on once Augur becomes popular?
There are a couple ways to mitigate this.  The primary technique is something called branching, basically, branching separates each event submitted on Augur into a subject category.  So let’s say we’re betting on science questions, and the general science category gets too big and reporters begin to be overwhelmed.  Then, someone can initiate a branch to separate science into categories like physics and not-physics.  If not-physics gets too big, it can be separated into branches like chemistry, biology, computer science, etc.  
Another way to remedy this problem is to only require a subset of reporters to report on certain events.  Let’s say there are 5,000 events, you would then have a chance to report on each one, such that you only had to report on something like one or two-hundred. 

There seems to be too much time delay from the time of closing an event to refereeing on it. It seems that this process might take between a month or more, so if I was a customer (predictor) for example and I win my prediction it will take me up to a month to get my winnings. This seems like too long in my opinion.

After an event happens you can sell immediately. Since the event has already happened, people who lost will sell for as much (aka little) as they can get and people will buy the correct outcome if it's <$1 - fees (because on official resolution they'll get $1). So this pushes the correct outcome price near 1. Then people who want to can sell their shares immediately. They do pay another trading fee, but since trading fees will be logarithmic (aka bigger near even odds and smaller at the edges) it'll be pretty small. So the reporting system is really useful for 2 things: 1) in a bush-gore scenario, to determine the outcome once the knowledge is known and 2) to report on outcomes, ensuring that there's a forceful incentive for the above to occur. The oracle system makes sure that the markets have an incentive to behave the way they do.  You can sell after an event has happened before the official referee payout is determined, but will pay a small trading fee. People who want to profit off the small interest between then will buy winning shares as long as they're <$1-fees.

Do I have to report on all events?
There's a few ways to scale the system (or we could even combine them)
  1. have a minimum number of events that people have to report on, anything beyond that they can choose to or not
  2. select a random subset of events for each reporter to report on
  3. encourage branching (aka separation into separate subject categories - i.e. elections, weather, etc.).
1 is the most doable (as in, it would definitely work & doesn't cause any major problems for the system), 2 is a bit more of a challenge, and 3 would definitely work as well.
Either way you're not reporting on all events.

We want it to take around a couple hours / two months to report - 200 seems about right for a minimum #, but we won't know for sure until we  get some actual real life scenarios when beta testing (our alpha was more focused on making markets, buying/selling shares, etc).
Will you allow for multiple markets on the same topic? "For example, you can make "Donald Trump will lose the Presidential Election" and try to market your Prediction Market to have more traction (investors/bettors) than mine?"
You can make multiple markets about the same thing --- which is good for reporters because it means more fees.  So if a market maker chooses a 2% fee, 1% goes to reporters, 1% goes to market creator each trade.  But it doesn't actually make more work for reporters because all the similar Donald trump markets can use the same Donald trump event
So the market maker needs to provide liquidity to start a market, i.e. equal to what people will bet on one side or another until you get people adding in money on both sides, correct? So really the market maker will 'make the first bet' in the Prediction Market he sets.
Yep!
"I would assume that REPs will receive a % of the value of the Prediction Markets that they confirm. However ... what would that look like in general terms?"
In general terms people who create markets can set their trading fees.  Half goes to the market creator, half goes to reporters.  There'll be a minimum set of events that people need to report on (we haven't decided on the exact # yet because we want to see how long it takes to report in practice, but we want it to take ~2hr or so every two months).  Trading fees from events will be split up amongst people who report.  So beyond that core group of minimum events if you don't report on other events you wouldn't get the fees from them.
There is an alternative way to scale things, which entails having reporters report on a randomly selected subset of events, but this would change the security model by a good amount and not enough research has been done into it.  So for now the above is how we're looking at scaling things.