Kleros is a decentralized arbitration protocol — a "court" that resolves the subjective disputes smart contracts cannot decide on their own: was a freelance deliverable actually delivered, is a token listing legitimate, did an insured event really occur. Its name and design borrow from the Athenian kleroterion, the stone machine that selected citizen jurors by lot. Where a price-feed oracle like Chainlink reports objective data, Kleros adjudicates judgment calls — and it does so as a DAO, where the same token that pays jurors also governs the protocol.

The court: stake PNK, get drawn by lot

Anyone can become a juror by staking PNK (pinakion) in a court. When a dispute arrives, the protocol draws jurors at random from the staked pool, weighted by stake: the more PNK you stake, the higher your probability of being drawn for any given case (buying a controlling share of the vote therefore costs progressively more as PNK grows scarce). Drawn jurors review the evidence submitted by both parties and vote on the ruling — typically Yes, No, or Refuse to Arbitrate — often through a commit-and-reveal scheme that hides each vote until all are cast, so jurors cannot simply copy one another. Where the PNK sits differs by version: in V1 it never leaves the juror's wallet and is simply locked against the subcourt, while V2 transfers it to the contract.

Coherence: making honesty the focal point

Kleros never tells a juror what is "true." Instead it pays jurors to converge on what they expect other honest jurors to pick — a Schelling-point game drawn straight from coordination theory. After a ruling, jurors who voted coherently (with the majority) collect the arbitration fees plus PNK redistributed from the jurors who voted incoherently; incoherent jurors forfeit a portion of their staked PNK. A juror who votes randomly, or is bribed to vote against the evident truth, is more likely to land in the minority and lose money — so the profit-maximising move is to read the evidence and vote the honest answer. This is the same "vote for what you think others will honestly vote" logic that secures UMA's DVM, implemented as a persistent, subcourt-structured jury rather than a single dispute round.

The court tree and appeals

Courts form a tree: a General Court at the root, with specialized subcourts beneath it (e.g. curated-list, oracle, or commerce disputes), each with its own parameters, stake requirements, and juror pool. A dispute can be appealed: each appeal round draws a larger jury — roughly twice the previous size plus one — and both sides (or third-party crowdfunders, who are rewarded if their side wins) must fund the higher juror fees. Because winning an appeal requires corrupting an ever-larger, freshly-drawn jury, the cost of buying a wrong outcome rises with each round, while an honest litigant can always escalate. If the protocol itself were ever 51%-attacked, PNK's final backstop is a governance fork that removes the attacker's stake — a recourse an external token could not provide.

Court V2: the same game, rebuilt modular on Arbitrum

Kleros has been running two courts in parallel since 2024. Court V1 is the production protocol described above: a single monolithic contract, KlerosLiquid, deployed on Ethereum L1 and Gnosis Chain, which has resolved more than 1,500 disputes since 2018. Court V2 runs on Arbitrum L2 and, in Kleros's own words, “keeps the same crypto-economic model as V1 and changes how it is implemented and where it runs.” The monolith is split into KlerosCore (dispute coordination), a separate SortitionModule (juror drawing), and pluggable DisputeKits that supply the resolution mechanism itself.

Three of those changes matter to anyone reasoning about the design rather than the code. First, the vote-hiding scheme became a choice: V1 has one mechanism (commit-reveal with plurality aggregation), while V2 ships Classic, Shutter (threshold-encrypted rather than commit-reveal), Gated (eligibility gated by a soulbound token) and GatedShutter — so a subcourt can pick how secret and how permissioned its jury is. Second, cross-chain disputes are relayed rather than redeployed: V1 meant separate, unconnected deployments per chain, while V2 routes disputes from any EVM chain to the Arbitrum court over the VEA bridge. Third, staking custody moved — PNK is transferred into the contract instead of being locked in the juror's own wallet. Running the court on an L2 is the point: the drawing, voting and appeal rounds described above are gas-heavy, and they were priced out of small disputes on L1.

V2 is still in beta — 100+ disputes processed, with internal review and external audits ongoing — and V1 remains live and paying staking rewards, so the two are not a migration so much as an overlap. (Checked 8 September 2026.)

PNK as a governance token, and the wider stack

PNK's second job is governance. Holders propose Kleros Improvement Proposals (KIPs), debate them on the forum, and vote via Snapshot; approved parameter changes are executed on-chain by the Governor contract, so the DAO can add courts, retune stakes and fees, or alter the very token supply. On top of the court sit Kleros's products — Escrow (arbitrated P2P payments), Curate (token-curated registries whose entries are policed by the court), Proof of Humanity (a sybil-resistant registry of real people), and a general-purpose arbitrable-transaction oracle that any dApp or DAO can plug into for a credibly-neutral verdict.

How Caper approaches this

Kleros exists because collectives constantly face questions that no vote can settle honestly on its own — "did this actually happen?", "is this claim legitimate?" — and someone must adjudicate them without a trusted middleman. Caper does not run a court, and it would be dishonest to pretend it arbitrates disputes. Its answer to the deeper problem Kleros targets — what do you do when the group reaches a decision you consider illegitimate? — is not adjudication but exit. A caper member who believes the DAO is being run against their interests can redeem their proportional slice of the treasury and leave, because a member's vote weight is the same canonical share that governs their exit payout. Kleros makes defection costly by paying jurors to converge on the truth; Caper makes capture costly by guaranteeing that anyone outvoted can always walk away with their fair share intact. Both are Schelling-point designs — one coordinates jurors on an honest verdict, the other coordinates members on a proportional, non-dilutable claim.

References

• Clément Lesaege, Federico Ast & William George, Kleros Short Paper — the Schelling-point juror model, appeals, and coherence incentives.
• Clément Lesaege, William George & Federico Ast, Kleros Long Paper (Yellow Paper) — full cryptoeconomic specification.
• Kleros docs, PNK Token — 915,528,222 total supply, matching the mainnet contract's on-chain totalSupply (the 764,626,704 figure still printed on the docs' Tokenomics page is stale); juror-staking and governance roles, 51%-attack resistance.
• Kleros docs, Court — drawing, evidence, Yes/No/Refuse-to-Arbitrate voting, and the court tree.
• Kleros docs, Appeals — appeal-fee crowdfunding and the asymmetric 2× / 1× funding rule that decides an unanswered appeal by default.
• Kleros blog, Governor Explainer; KIP-69 Parameter Updates (Apr 2024) — on-chain governance via KIPs and Snapshot.
• Kleros docs, Court — Overview — the V1/V2 comparison table, the DisputeKit list, the VEA bridge, and the beta status of Court V2 on Arbitrum.