How Stable Are LLM-Generated Proofs Under Program Refactoring? A Stress Test for Verifiable Code Generation
RefactorProof measures unchanged-proof stability under program refactoring, not the cost of maintaining a proof after it breaks, which derives 458 Lean-certified variants from the 185 VERINA tasks that compile in the authors' pinned environment while holding the specification and evaluated proof fixed byte-for-byte.
A. Mandiga, Guo-Ming Li
· 0 citations