Spec-Harness: Measuring and Improving Behavioral Adequacy of LLM-Synthesized Formal Specifications
Comparisons of classical and prompt-based JML synthesis approaches are compared, and it is found that prompt optimization through verification feedback raises pass rates but reaches a clear ceiling, and Spec-Harness is introduced, a framework that measures the behavioral adequacy of a specification along four dimension...
M. Rakib, Hossain Misu, IrisMa et al.
· 3 citations