research_project · DiggingBeagle record
SolBehave benchmark
SolBehave is a 2026 research benchmark from Bologna and Copenhagen researchers for checking whether LLM-regenerated Solidity contracts preserve behavior across differential fuzzing, historical transactions and synthesized stateful sequences. Its primary paper uses 250 source-verifiable Ethereum contracts and seven open
- Entity kind
- research_project
Record description
SolBehave is a 2026 research benchmark from Bologna and Copenhagen researchers for checking whether LLM-regenerated Solidity contracts preserve behavior across differential fuzzing, historical transactions and synthesized stateful sequences. Its primary paper uses 250 source-verifiable Ethereum contracts and seven open-weight models. The code repository is public at the time of this review, although its README contains conflicting statements about dataset availability; no independent benchmark rerun was performed.
The 250-contract corpus follows reproducibility and size filters. The behavioral tests are complementary: differential fuzzing stresses function inputs, historical replay preserves observed use, and DCR-guided sequences exercise additional valid call orderings. A passing test does not certify total semantic equivalence.