Skip to content

brief.md

Latest commit

 

History

History
139 lines (104 loc) · 6.08 KB

brief.md

File metadata and controls

139 lines (104 loc) · 6.08 KB
Raw

Kleros: The Oracle for Prediction Markets

A technical brief on Kleros as the resolution layer for prediction markets. This is the markdown version of the interactive brief at https://oracle.kleros.io/.

1. The resolution layer

The oracle for prediction markets. A prediction market is only as trustworthy as the layer that resolves it. When two sides disagree on an outcome, Kleros settles it through a neutral, transparent due process that scales alongside the platform.

  • Scalability — A randomly drawn jury per case, from a subcourt suited to the topic. With 750+ jurors already staked, capacity is not a constraint.
  • Due process — Structured evidence, reasoned votes, and appeals that double the jury each round.
  • Credible neutrality — Contested outcomes decided by an independent, randomly selected jury.

2. Proof, before the pitch

Live since 2018. Proven in production.

  • 2018 — Running a live, on-chain arbitration system since.
  • 2020 — Resolving prediction-market disputes. Omen (now Presagio) was first to integrate.
  • 2,653 — Disputes resolved on-chain across Ethereum, Gnosis & Arbitrum.
  • +750 — Jurors participating in Kleros arbitration.

With Reality.eth: five years of contested outcomes, settled by jury.

3. How it works

One question. A two-layer resolution.

Reality.eth, the optimistic layer:

  1. Question posted — with a bond, naming Kleros as the arbitrator.
  2. Bond escalation — anyone answers or challenges by doubling the bond. A wrong bond is forfeited. Most questions settle right here; no court needed.

On genuine disagreement, anyone pays a fee to escalate to Kleros Court, the arbitration layer:

  1. Sortition: jurors drawn — randomly, weighted by staked PNK, from the relevant subcourt.
  2. Vote: evidence & ruling — jurors vote independently. Coherent votes are rewarded; incoherent ones slashed.
  3. Settle: back to the market — the ruling flows back to Reality.eth and settles the market.
  4. The appeal escalator — jurors per round: 3 → 7 → 15 → 31 → … Each appeal doubles the jury, so overturning a well-evidenced answer becomes progressively more expensive.

Jurors cannot be chosen, and they cannot choose their case. More PNK raises your probability of being drawn, but it never buys the seat.

4. Why not UMA

Same game. Different table. Both reward voters for converging on the answer the panel is expected to choose, a Schelling point. The difference is who sits on the panel.

UMA Kleros
Who answers The entire token-holder base votes on every dispute. A fixed-size jury, drawn per case from a topic-relevant subcourt.
Who decides Token-weighted vote of holders. A random, stake-weighted jury draw, per case.
Scalability Everyone votes on everything. As volume climbs, voters can't study each case. Each case gets a dedicated jury drawn from 750+ active jurors, so review depth holds as volume grows.
Due process No appeal; escalation to the DVM, where ~9 wallets hold 50% of voting power. Evidence + appeals. The jury doubles each round, raising the cost to overturn a coherent answer.
Speed ~2 days. Most cases resolve in a day; heavily appealed cases can take a couple of weeks.

5. On real money

Four disputes, settled by a drawn jury. Every ruling is a live, on-chain link.

  • Case #532 · Omen/Presagio — "Will Joe Biden win the 2020 US presidential election?" General Court, 500 votes. Ruled Yes.
  • Case #302 · Omen/Presagio — "≥1,000 reported US COVID deaths in a single day, first half of July 2020?" General Court, 500 votes. Ruled Yes.
  • Case #1661 · Seer — "Was Zelenskyy photographed or videotaped wearing a suit between May 22 and June 30, 2025?" General Court, 31 votes. Ruled Yes.
  • Case #1657 · Seer — "Winner of the run-off in the 2024 Romanian presidential election?" General Court, 31 votes. Resolved too soon, run-off annulled.

The hard cases, including the one where the right call was that the market resolved too soon.

6. When oracles fail

Not every contested outcome is an oracle failure. Some contested Polymarket outcomes trace to the market's own rules or a late clarification, not the oracle. But the oracle layer can be a structural weak point in its own right.

On UMA, voting power is concentrated in a handful of large holders, and a majority of active voters are linkable to Polymarket trading accounts, a direct positional conflict in several markets.

Under Kleros: 25% of stake is a 25% chance of being drawn as a juror, not 25% of the verdict. And every appeal raises the cost of holding a dishonest result, making concentration expensive rather than decisive.

7. Who's already in

Prediction markets resolving with Kleros:

  • Seer — Decentralization, composability, and community-driven resolution.
  • Omen / Presagio — Built on Gnosis' Conditional Tokens Framework. First to integrate, 2020.
  • Foresight — Kleros's own forecasting experiment, running on Seer.
  • Precog — Collective intelligence for human and AI participants.
  • Futarchy.fi — Prediction markets & conditional tokens for DAO decision-making.
  • Deep Funding — Allocation proposals run through conditional prediction markets.
  • Ringo — Open markets on real-world events, integrated with X.

8. Integrate

Make Kleros your resolution layer. Plug Reality.eth and Kleros into your market in a few steps. Optimistic by default, with a credible court behind the hard cases.

Available on Ethereum, Arbitrum, Gnosis, Polygon, Base, Optimism, Unichain. Solana integration in progress.

Contact: juanrah@kleros.io · https://t.me/juanrah

Vitalik's take on Kleros