Skip to content

Prior Art

Fact-only comparison of Accord against the systems that share its problem space. There is no winners column — the dimensions are chosen so a reader can locate each system in the design space and draw their own conclusions. Every cell describes documented behavior of the named system, cited to a primary source. Facts age: this page was written 2026-09-28; re-verify cells against the cited sources before relying on them.

Systems compared (the ones with living, citable primary documentation):

  • Kleros — crowdsourced juror court on EVM chains, live since 2019. Accord's acknowledged lineage.
  • UMA — optimistic oracle with escalation to a token-holder vote (the DVM). The other Schelling-lineage design.
  • Kourt — prediction-market courts on gno.land; open coin-weighted electorates. Included because its design documentation is the most adversarially honest in the space and exercises a genuinely different cell of the design space (no sortition at all).

Aragon Court is deliberately absent: its documentation is currently offline and we will not state uncited cells. It was a first-generation Ethereum court (drafted jurors, commit-reveal, final appeal to all jurors); Kleros covers that region of the space with living documentation.

The table

Dimension Accord Kleros UMA (OOv2/v3 → DVM) Kourt
What it is Arbitration oracle: any Solana program files a dispute via CPI, gets a ruling (ADR-0004) Arbitration protocol + product suite (Curate, Escrow, Proof of Humanity) (docs) Optimistic oracle for data/assertions; disputes are the exception path, not the product (docs) Prediction-market courts; the permanent record of contested claims is the product (whitepaper)
Platform Solana (Anchor, sBPFv3) EVM (Ethereum, Gnosis Chain, …) EVM gno.land (Gno realms)
Decision body Drawn jury: stake-weighted sortition from a per-pool accumulator Drawn jury: PNK-stake-weighted sortition per court All UMA token holders who vote the request (broadcast) All court-coin holders (broadcast)
Selection / randomness VRF committed before the draw; accumulator root frozen atomically with it; per-seat on-chain-verifiable sortition (ADR-0012) Sortition over staked PNK by court (sortition) n/a — no selection; weight is token balance at vote n/a — weight is coin balance, sealed at a pre-vote epoch
Vote secrecy Commit-reveal with per-juror salt hash Hidden votes (commit-reveal) in v2 (docs) DVM: committed votes, revealed at end Open ballots; dissent is preserved on the record
Who pays jurors The filer (per-juror fees, posted at filing) The filer (arbitration fees) Proposers/disputers post bonds; voters earn from rewards/slashing Minted court emission (bounded, decaying); explicitly never counterparty money (regulatory posture)
Loser penalty Slash α·min_stake of incoherent jurors' stake, redistributed to coherent (stake mint) Incoherent jurors' PNK is partially slashed and redistributed (docs) Bonds forfeited on the wrong side of a resolved dispute Forfeited bonds are burned — no one is paid from another's loss (whitepaper §2)
Appeals / escalation Permissionless appeal to 2N+1 panel, rising bond + flip bounty (ADR-0030) Anyone can fund an appeal; rounds grow ~2× (docs) Dispute escalates to the DVM token vote; one escalation layer No merits appeal — verdicts are never revised; re-ask as a new claim. A Review Court hears moderation appeals only
Finality Final ruling retro-settles every round (ADR-0018) Ruling executes when appeal window lapses DVM resolution is final Verdict recorded, never revised; the record is the product
Conflict-of-interest handling Priced, not excluded: a drawn party-juror is slashed if incoherent (accepted residual — trust profile) Policy-level (juror guidelines); not enforced in-protocol n/a (no jury) Hard exclusion: stakers/author/answerer cannot vote; the record survives withdrawal (whitepaper §2)
Weight-freeze vs. buy-in attacks Accumulator root frozen at VRF commit; drawn seats locked (active_draws) until settlement Stake snapshot per draw period Token snapshot at vote min(sealed-epoch weight, live balance) + coin locked until resolution (RENTEDWEIGHT)
Low turnout / no-show Reveal quorum (2/3) + tie ⇒ same-size redraw ladder, no-shows slashed; exhaustion ⇒ Failed with exact refunds (ADR-0021, 0033) Ruling by drawn votes cast; appeal is the corrective No-dispute (optimistic) default; unresolved dispute escalates Quorum floors keyed to claim size; failed rounds burn windows; unanswered claims expire with refunds
Pool structure Permissionless Subaccords, each with own staking + fee mint (ADR-0020) Court tree (general → specialized), stake per court Single global DVM One court per topic, each with its own bonding-curve coin
Parameter governance Per-Subaccord authority, 48h timelock; program upgrade via multisig → freeze (ADR-0005, 0007) PNK governor + court parameter votes UMA token governance (UMIPs) Per-court governor over checkpointed coin votes; meta court over moderation
Evidence confidentiality On-chain hash + trusted re-encryption operator (ADR-0006, 0015) Public evidence standard (ERC-1497) Public assertions + ancillary data Public claims + public record

Design-space notes

  • Sortition vs. broadcast is the load-bearing split. Accord and Kleros draw a small jury and pay it from filer fees; UMA and Kourt broadcast the question to a token-weighted electorate. Sortition makes turnout structural (the drawn must respond or the round redraws); broadcast must solve turnout (quorum floors, emission-paid voting) and weight-rental attacks (Kourt's min(snapshot, live) + vote-lock is the most documented treatment).
  • Counterparty-funded vs. minted incentives. Accord and Kleros fund coherence by slashing incoherence — the incentive is the transfer. Kourt deliberately refuses that shape (minted rewards, burned bonds) to remove wager substance. UMA sits between (bonds at stake, token rewards for voting).
  • Appeal philosophy. Accord escalates panels and retro-settles every round against the final ruling; Kleros escalates similarly; Kourt pins finality at first verdict and treats re-asking as a feature (claims are wordlocked, so re-asking produces comparable record entries). Different products: an escrow arbitration wants convergence, a record wants permanence.
  • Where Accord sits. Closest to Kleros in mechanism (drawn jury, coherence slashing, permissionless appeal), differentiated by: live on-chain stake accumulator with no snapshot poster (the Kleros analog trusts a drawn-from stake registry at draw time), two-mint economics separating collateral from compensation, the no-ruling-no-pay refund contract for Arbitrables, and Solana-native composability (two-CPI integration).

Sources