Benchmark RadarMPC-Patch-Bench
MPC-Patch-Bench evaluates LLM-based code repair on repository-level Secure Multi-Party Computation (MPC) software. It provides 205 verified instances with Fail-to-Pass/Pass-to-Pass tests and a verifier that checks cryptographic safety and numerical fidelity via differential testing and static analysis.
- Released
- 2026-06-09
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Existing benchmarks lack MPC-aware evaluation for repository-level code repair. MPC-Patch-Bench addresses security and numerical-fidelity gaps, offering a repeatable protocol for assessing LLM agents on real-world MPC tasks, with verification rejected up to 40% of functionally passing patches.
Motivation
Repository-level benchmarks for evaluating Large Language Model (LLM) code repair on Secure Multi-Party Computation (MPC) software do not yet exist, and directly transplanting general-purpose benchmarks such as SWE-bench fails on three structural fronts: (i) MPC repositories are dominated by generic Python infrastructure rather than cryptographic logic; (ii) high-value MPC fixes lack the standardized tests rigid ext…
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.