Local Success Does Not Compose: Benchmarking Large Language Models for Compositional Formal Verification

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Xu, Xu, Li, Xin, Qu, Xingwei, Fu, Jie, Yuan, Binhang
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912611400941568
author Xu, Xu
Li, Xin
Qu, Xingwei
Fu, Jie
Yuan, Binhang
author_facet Xu, Xu
Li, Xin
Qu, Xingwei
Fu, Jie
Yuan, Binhang
contents We introduce DafnyCOMP, a benchmark for evaluating large language models (LLMs) on compositional specification generation in Dafny. Unlike prior benchmarks that focus on single-function tasks, DafnyCOMP targets programs composed of multiple interacting functions with data dependencies, requiring reasoning across component boundaries. The benchmark consists of 300 automatically synthesized multi-function programs. We evaluate several state-of-the-art LLM families and find that, while they perform well on single-function verification, their performance drops sharply on compositional tasks. Analysis reveals systematic failures in cross-functional reasoning, including fragile specifications, misalignment between implementations and proofs, and unstable reasoning. DafnyCOMP thus provides a diagnostic tool for measuring progress toward reliable, verifiable, and compositional code generation with LLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2509_23061
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Local Success Does Not Compose: Benchmarking Large Language Models for Compositional Formal Verification
Xu, Xu
Li, Xin
Qu, Xingwei
Fu, Jie
Yuan, Binhang
Programming Languages
Artificial Intelligence
We introduce DafnyCOMP, a benchmark for evaluating large language models (LLMs) on compositional specification generation in Dafny. Unlike prior benchmarks that focus on single-function tasks, DafnyCOMP targets programs composed of multiple interacting functions with data dependencies, requiring reasoning across component boundaries. The benchmark consists of 300 automatically synthesized multi-function programs. We evaluate several state-of-the-art LLM families and find that, while they perform well on single-function verification, their performance drops sharply on compositional tasks. Analysis reveals systematic failures in cross-functional reasoning, including fragile specifications, misalignment between implementations and proofs, and unstable reasoning. DafnyCOMP thus provides a diagnostic tool for measuring progress toward reliable, verifiable, and compositional code generation with LLMs.
title Local Success Does Not Compose: Benchmarking Large Language Models for Compositional Formal Verification
topic Programming Languages
Artificial Intelligence
url https://arxiv.org/abs/2509.23061