SI-RVP: A Hybrid Optimistic Rollup Dispute Protocol With Off-Chain Bisection and On-Demand Single-Instruction Validity Proofs

초록

Optimistic rollups secure Layer-2 execution behind a multi-day fraud-proof challenge window-typically seven days-that delays withdrawal finality and locks user capital. Existing dispute mechanisms either keep the bisection game and single-step execution fully on-chain (as in Cannon and BoLD), incurring a cost of several million gas, or replace it with a full-batch validity proof (as in Morph and Kailua), shifting proving cost onto every batch. We present SI-RVP (single-instruction responsive validity proof), a hybrid optimistic-rollup dispute protocol that combines off-chain bisection, a Poseidon-aligned bisection commitment, an on-demand single-instruction Groth16 proof, and a permissionless validator pool that remains secure as long as at least one validator is honest. We provide a structural correctness argument: safety reduces to Groth16 soundness over an explicit single-step execution relation together with hash collision resistance; liveness bounds every dispute in which both parties follow the protocol to a logarithmic number of off-chain rounds plus four sequential on-chain transactions, while a stalling party is forced on-chain by an attributable force-move rule at a logarithmic number of moves; and soundness is established for the specified constraint system. A Sepolia proof-of-concept measures the on-chain cost of a fully played dispute at approximately 0.83 million gas over its five-transaction lifecycle, where a fully played Cannon dispute requires 73 sequential bisection moves plus one final leaf re-execution - about 74 on-chain transactions in total, the last of them multi-million-gas. The adversarial path thus reduces, structurally, to four sequential Layer-1 confirmations plus a logarithmic off-chain bisection; under an assumed mainnet-finality confirmation of 10 to 12 minutes per transaction (a modeling assumption, not a measurement), this evaluates to roughly 47 minutes of dispute-resolution latency. The challenge window itself is retained as an availability buffer, so finality in the no-dispute case is unchanged. The fallback probability admits a simple closed form in the validator-pool size and the per-validator honest-participation rate, under an independence model whose sensitivity to shared infrastructure we quantify separately. All contracts, circuits, and measurement artifacts are public. Production deployment requires a multi-party trusted-setup ceremony, which we identify as the principal deployment-blocking dependency.

키워드