Kleros

Kleros is a decentralized arbitration protocol on Ethereum – a general-purpose “crypto supreme court” that renders binding rulings on subjective disputes that a smart contract cannot decide by itself. When code needs a human judgement – did a freelancer deliver the work, is this token listing legitimate, is this a real person – Kleros supplies a decentralized, game-theoretic answer instead of a trusted human moderator. It is the most widely deployed of the DAO-tooling category's dispute-resolution primitives, and a common building block behind curated registries, escrow, and identity systems.

The gap Kleros fills: disputes code can’t settle

On-chain governance is good at questions with objective, verifiable answers: did a proposal reach quorum, did a transfer execute. It is helpless at subjective questions – whether delivered work met a spec, whether a submitted photo is really the claimed person, whether an insurance condition was met. Traditionally these route to a trusted moderator, an admin multisig, or an off-chain court, each of which reintroduces exactly the centralized chokepoint a DAO is trying to remove.

Kleros positions itself as a neutral third party for these cases: an arbitration protocol that any contract can call to obtain a ruling, and whose jurors are anonymous, self-selected, and economically motivated to answer honestly. It is the decentralized analogue of an appeals court – invoked when a smart-contract relationship needs judgement rather than computation.

The court: staking PNK and random selection

Jurors opt in by staking PNK (pinakion) in a subcourt – a specialization such as general, software development, or curation. Courts form a tree with the General Court at the root; staking in a subcourt automatically stakes you in every court above it. When a dispute arises, jurors are drawn at random with probability proportional to the PNK they have staked: the more you stake, the more often you are selected, which also puts more of your own capital at risk on each case.

Drawn jurors review the evidence and cast a vote – typically Yes, No, or Refuse to Arbitrate for invalid or illegal cases – and, in hidden-vote courts, first commit a hashed vote and reveal it later so jurors cannot copy one another. Jurors are expected to justify their vote in writing. Neither party knows who the jurors are, which is what makes the panel hard to bribe or intimidate.

The coherence incentive: a Schelling point for truth

Kleros never asks a juror to prove they are right – it rewards coherence with the eventual majority. Jurors who vote with the majority are paid the arbitration fees the parties deposited, plus PNK redistributed from jurors who voted incoherently. Jurors on the losing side forfeit part of the PNK they staked to be drawn.

This turns the honest answer into a focal (Schelling) point: absent collusion, a rational juror’s best guess of what other jurors will conclude is the true, evidenced answer, so voting honestly is the profit-maximizing strategy. A juror who votes randomly without reading the evidence is more likely to land incoherently and lose their stake, so over time careless or dishonest jurors bleed capital and leave. The security model is economic, not reputational – influence and risk both scale with money staked.

Appeals: escalating juries and crowdfunded fees

Any party who disagrees with a ruling can appeal, which re-tries the case before a larger panel – the number of jurors roughly doubles each round, raising both the cost of capturing a majority and the fees at stake. Because appeal fees can be crowdfunded, third parties who believe a ruling is wrong can chip in toward the correct side’s appeal, and crowdfunders backing the side that ultimately wins are financially rewarded. The escalating cost is deliberate: it makes frivolous appeals expensive while giving an obviously-wrong first-round result a path to be overturned by a bigger, more expensive jury.

Beyond the court: Curate, Escrow, Proof of Humanity, Scout

Kleros Court is the engine; most people meet Kleros through the products built on it:

• Kleros Curate – token-curated registries: decentralized lists (of tokens, addresses, content) where entries are challenged and disputes settle in court. This is the modern descendant of the token-curated-registry idea.
• Kleros Escrow – a payment escrow where a disputed release is arbitrated by the court rather than a platform.
• Proof of Humanity – a Sybil-resistant registry of unique humans (video submission plus social vouching, issuing soulbound IDs), with challenges adjudicated by Kleros. It underpins UBI-style distribution and one-person-one-vote experiments.
• Kleros Scout – a community-curated database of contract and token metadata used to flag scams inside wallets and explorers.

Each product is really the same pattern: put a subjective claim on-chain, let anyone challenge it, and fall through to the court when they do.

The PNK token, governance, and V2

PNK does double duty: it is the stake that gates jury selection (and the collateral slashed for incoherence) and the governance token of the Kleros DAO, which votes on protocol parameters, court creation, and treasury. Notably, PNK is deliberately not a general-purpose payment token – arbitration fees are paid in ETH or stablecoins, while PNK is the skin-in-the-game that secures juror honesty, an intentional separation so that buying influence over rulings is distinct from paying for them.

Kleros V2 re-architects the court and extends it beyond Ethereum mainnet to additional chains, lowering the gas cost of participating; current deployments are listed in the developer docs.

Limits and failure modes

Kleros’s guarantees are probabilistic and economic, and they have edges. A 51% attack is possible in principle: an actor holding a majority of the PNK staked in a small subcourt can force coherent-but-wrong verdicts, which is why high-value disputes belong in deep, well-staked courts and why appeals escalate. P + ε bribery – paying jurors a small premium above the honest payoff to vote a certain way – is the classic theoretical attack on Schelling-point schemes, mitigated but not eliminated by appeals and stake redistribution. Rulings are only as good as the evidence and policy the parties and subcourt supply, so a poorly specified dispute yields a poorly reasoned verdict. And because influence tracks staked PNK, the court’s neutrality rests on token distribution being wide enough that no single party can quietly dominate a subcourt.

How Caper approaches this

Caper has no arbitration court, and its core decision – how a caper’s treasury is spent – is settled in two phases rather than by a jury. A proposal names up to five ranked options, one of which must be a “Do nothing”, and banks a fee at creation – 1% of the largest option's XRD value, at least 500 XRD – that the treasury keeps whether the proposal passes or fails (PROPOSAL_FEE and PROPOSAL_FEE_RATE, contracts/logic/src/lib.rs). Token holders then cast ranked ballots at 100 XRD each, and a permissionless crank folds them into a Borda tally that passes an option only if its share of the weight actually cast clears a supermajority of 1.5 / option_count – 75% on a two-option ballot, 30% on a five-option one.

A legislative pass is not execution; it opens the second phase. trigger_proposal locks the caper’s trailing time-weighted average price as a baseline and opens the market window, and when that window closes resolve_proposal compares the TWAP over the window itself against the baseline, executing only if the market held. Integrating the window rather than reading spot at the crank-second is deliberate: a displacement has to be held for a real share of the window, so an atomic buy→resolve→sell round trip buys no verdict.

That makes the two protocols closer than they look, and sharpens the contrast where it survives. Kleros prices honest adjudication by putting purchasable PNK stake at risk; Caper prices dissent by making the objection an actual sale. A member who disagrees calls exit() and redeems their fair share of the treasury, governed by the same earned weight their ballot carried – and that sale pushes the price down, which is precisely the quantity the pending window is measuring. The exit right is not an alternative to the verdict here; it is an input to it.

One correction worth stating plainly, because it reverses what this page said until the genesis redeploy of 11 September 2026: v, the soulbound half of that weight, is purely a voting record. It is minted on exactly one member-facing surface – casting a ranked ballot mints one v (VOTE_MINT, contracts/logic/src/lib.rs) – and buying, selling and transferring mint none. Until that redeploy trades minted it too, at 0.01 per XRD of gross value on each leg, which made v a record of XRD put at risk on whichever surface you chose; it is now a record of having voted, and of nothing else.

It is soulbound only in the transfer sense: its depositor role is bound to the caper’s own state component with the updater set to deny_all, so it cannot move between wallets (contracts/core/src/caper_dao.rs) – but it can certainly be acquired, by anyone, at a published price: the flat 100 XRD a ballot costs (VOTE_FEE). The two protocols remain complements rather than competitors: a caper that needed a genuinely subjective fact settled – did a funded milestone actually ship – could read an oracle like Kleros for that fact while keeping the money decision on its own rails.

References

• Kleros, kleros.io and documentation.
• Kleros docs, Court: how it works, Appeals and Tokenomics (dispute lifecycle, appeal-fee crowdfunding, the coherence incentive, PNK).
• Kleros, Court, Curate, Escrow, Proof of Humanity.
• Kleros, Deployment addresses (V2 chains).
• Kleros, GitHub organization and blog.