Kleros Research Roadmap

Scientific challenges for the future of decentralized justice…

At Kleros we are doing exciting work to apply cryptoeconomic reasoning and game theory to solve practical problems which will positively impact society as a whole.

To give our community more insight into the kinds of challenges we are facing, we have decided to share a rough roadmap of our research projects and objectives for the mid to long-term.

The theory behind Kleros is already pretty well developed in the case where jurors have to decide between two options. We have a prototype that is built based on these ideas.

A video walkthrough of Kleros prototype. This version is built for cases where jurors have to decide between two options.

Our long term vision is that Kleros will be capable of solving a wide variety of disputes. When writing a smart contract that designates Kleros as its arbitrator, a user should be able to provide a range of different outcomes for jurors to vote. These outcomes might include giving one party more time to complete a project, or issuing a partial or full refund.

Furthermore, our proof-of-concept is just now in the process of moving from having a single court where all cases are decided to having families of subcourts specialized by type of dispute. In the long-term, Kleros governance mechanism will allow each of these subcourts to adjust its fees, time limits and other parameters to meet the needs of specialized cases. This is how Kleros becomes a multipurpose arbitration system able to adjudicate disputes in a fast, affordable and transparent way.

With this vision in mind, we are working on a number of research topics.


Arbitration fees should give jurors a fair reward for their work. Compensation for work done in a very specialized subcourt requiring substantial effort and special skills should be higher than that for a court that doesn’t require such effort and skills.

Moreover, the level of the fees affects how attractive it is to be a juror, which then affects the price of the pinakion (PNK) token, which in turn affects how expensive it is for an attacker to try to buy up enough PNK to perform a 51% attack.

The prediction market platform Augur has some similar concerns and they include a system where the fees automatically adjust to the value at stake in their markets, designed so that it is never profitable to try to hijack the system via a 51% attack.

With Kleros, it is harder to estimate the total value at stake because people may use the protocol to arbitrate disputes involving non-monetary assets. However, we are considering ways that fees can adjust in such a way that provides maximum security and also allows for the subcourts to adapt to their individual needs.

(For an in depth comparison between Kleros and Augur, read this article).

Non-binary and even non-linear outcomes

If there is a dispute between Alice and Bob, a freelancer that she has hired, the jurors should certainly have the ability to pay Bob or reimburse Alice. But maybe they should also be able to give Bob more time to finish the work.

While this seems simple, it can pose subtle challenges. For example, one could imagine a system where you order the choices (so giving Bob more time is “between” paying him and refunding Alice). Then one could pick as the outcome the option that received the median juror vote. However, without a careful handling of the juror incentives you can have situations where the jurors are biased towards voting for outcomes in the middle as a hedge against being penalized.

In some cases - particularly in multi-party disputes - there may not even be a natural ordering on the outcomes. Think of a situation where Alice rents her apartment to Bob on Airbnb. When the apartment is damaged, the jurors have to decide whether Alice was responsible for the damages before Bob’s arrival and should pay for them herself, whether Bob should cover the cost of the damages, or whether the damage is the result of an accident that should be covered by Alice’s insurance.

We have begun to explore certain cases where we can handle these multi-outcome cases through binarization. Namely, the choice given to the jurors is transformed into a series of binary choices. Generally, we intend to build upon voting theory, which faces similar problems. We are particularly inspired by the ideas of Condorcet voting schemes.

Nicolas de Condorcet (1743–1794) was a French philosopher, mathematician and political scientist. He developed a voting system known as the Condorcet method which selects the candidate who would beat each of the other candidates in a run-off election. Some of our current research ideas are based on Condorcet’s work.


A key application of Kleros is to oracles. Kleros allows to bring knowledge of the real world (the correct party in some dispute) onto the blockchain. This can be leveraged to bring more general information on-chain.

We have recently been working on a model for how to implement a price oracle via Kleros. This could be used by contracts that need to know, for example, the cost of 1 ETH in USD. For the moment, our work uses the binarization ideas described above. Also, for now, our model depends on the presence of at least one “respondent” who submits an honest price without compensation, except for perhaps getting some of the deposits lost by malicious respondents. (Such a respondent could be motivated by self-interested reasons as might be the case for a respondent who is also invested in a contract that uses this price oracle).

Going forward, we will work on how to properly incentivize respondents to oracles. Also, we still have research to do to be able to handle oracles that can submit information that isn’t naturally ordered, such as a response to a prediction market on which of five candidates was elected president.

Moreover, we are considering whether there are other applications where, like in the case of oracles, one can make use of repeated calls to Kleros, organized in some useful way to solve some problem that doesn’t look at first glance like a dispute resolution problem.


Once we have a substantial number of real cases being judged by real jurors motivated by real financial incentives, it will be interesting to observe how participants actually act, how that conforms to game theoretical models, and what are the implications for Kleros.

We intend to organize experiments on test cases to see how users respond to bribes. We may offer bounties to actors who manage to have specific malicious results adopted in such test cases in order to gauge the robustness of the system.

Consider the specific example of p+epsilon attacks. We anticipate launching a p+epsilon attack against a test dispute to see how users react. Our appeal system already provides a level of defense against p+epsilon attacks as an attacker has to commit (and risk losing) large sums of money in subsequent appeals with large numbers of jurors.

As we further explore defenses against these attacks, we will consider one of the responses proposed to deal with p+epsilon attacks, which is a sort of coordinated mixed strategy, where jurors act in a way to drain the attacker for a maximal amount of funds. Organizing such a coordinated response poses both game theoretic and practical challenges. Nevertheless, for the case of our test p+epsilon case, we will also launch a contract that allows counter-coordination to organize itself, and we will observe if these measures are effective.

Zero-knowledge elements

Schelling points are heavily influenced by what information parties have access to. However, on Ethereum, information added to the blockchain is publicly visible, so controlling the information jurors have access to is a complicated problem.

This is already an issue that has needed to be addressed even for the most simple versions of Kleros: we have jurors vote via a commit and reveal scheme so that their vote is only visible on the chain once the voting period ends. This avoids jurors from being able to wait until other jurors have voted and then mimicking their vote to ensure consistency.

However, we might try to be more sophisticated with what kind of information the jurors see, particularly as more gas-efficient zero-knowledge primitives are being added to Ethereum. For one example, if there is an appeal, should jurors know the decision of the case in the lower court and who is appealing? We can imagine (possibly in certain subcourts) concealing this information. Then when it comes time for a party to appeal, they can use a ring signature from a new address to prove that they are a party to the dispute without revealing who.

The Strategy of Conflict. Originally published in 1960, this book pioneered the study of bargaining and strategic behavior in situations of conflict. In this book, Schelling introduced the concept of focal point, which many people know today as Schelling Point.

Arbitration fee insurance

We can imagine attacks where a very well-financed attacker attempts to appeal a decision until the opponent just doesn’t have the resources to continue to pay appeal fees. Then even though arbitration fees are reimbursed to the party who is eventually ruled correct (so if the correct party kept going she would eventually be reimbursed) a correct party may be forced to give up due to lack of resources.

As a solution to this problem, we imagine that one might be able to obtain a loan from a sort of insurer. This insurer covers the fees of parties that it judges likely to be ultimately correct. In exchange, the insurer receives a part of the penalty that the attacker will ultimately pay once they are ultimately judged to be wrong. This idea is similar to “After the event’’ insurance, which is typically designed for covering legal expenses in personal injury cases, and can be purchased after the injury has occurred (after the “event”). Also, the results of such an insurance framework wind up being similar to those of the “no win, no fee’’ agreements offered by some law firms in traditional civil disputes.

We will explore how to structure this in a way that is accessible for users; for example, the insurer could be organized as a DAO.

Building upon Ethereum scaling solutions

As different scaling solutions are developped for Ethereum, including Plasma, generalized state channels, Truebit, etc., we will think about how to adapt Kleros to best build on these advances in order to optimize gas costs and minimize delays while maintaining security.

Kleros and traditional arbitration

Bringing in ideas from game theory and blockchains allows us to approach dispute resolution from a radically new perspective.

Nonetheless, many of the nuances of traditional arbitration will certainly translate to our system. Hence we will seek out expertise in traditional arbitration, and we will explore how best the contributions of cryptoeconomics can be integrated with these existing ideas.

In short, we are just at the beginning of a long research agenda. Our long term vision is that Kleros will be capable of solving a wide variety of disputes in a fast, transparent and inexpensive manner. An exciting roadmap for applying cryptoeconomic reasoning and game theory to solve practical problems which will positively impact society as a whole.

Learn More

Join the community chat on Telegram.

Visit our website.

Follow us on Twitter.

Join our Slack for developer conversations.

Contribute on Github.