CLEVER: A Curated Benchmark for Formally Verified Code Generation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Thakur, Amitayush, Lee, Jasper, Tsoukalas, George, Sistla, Meghana, Zhao, Matthew, Zetzsche, Stefan, Durrett, Greg, Yue, Yisong, Chaudhuri, Swarat
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915571958808576
author Thakur, Amitayush
Lee, Jasper
Tsoukalas, George
Sistla, Meghana
Zhao, Matthew
Zetzsche, Stefan
Durrett, Greg
Yue, Yisong
Chaudhuri, Swarat
author_facet Thakur, Amitayush
Lee, Jasper
Tsoukalas, George
Sistla, Meghana
Zhao, Matthew
Zetzsche, Stefan
Durrett, Greg
Yue, Yisong
Chaudhuri, Swarat
contents We introduce ${\rm C{\small LEVER}}$, a high-quality, curated benchmark of 161 problems for end-to-end verified code generation in Lean. Each problem consists of (1) the task of generating a specification that matches a held-out ground-truth specification, and (2) the task of generating a Lean implementation that provably satisfies this specification. Unlike prior benchmarks, ${\rm C{\small LEVER}}$ avoids test-case supervision, LLM-generated annotations, and specifications that leak implementation logic or allow vacuous solutions. All outputs are verified post-hoc using Lean's type checker to ensure machine-checkable correctness. We use ${\rm C{\small LEVER}}$ to evaluate several few-shot and agentic approaches based on state-of-the-art language models. These methods all struggle to achieve full verification, establishing it as a challenging frontier benchmark for program synthesis and formal reasoning. Our benchmark can be found on GitHub(https://github.com/trishullab/clever) as well as HuggingFace(https://huggingface.co/datasets/amitayusht/clever). All our evaluation code is also available online(https://github.com/trishullab/clever-prover).
format Preprint
id arxiv_https___arxiv_org_abs_2505_13938
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CLEVER: A Curated Benchmark for Formally Verified Code Generation
Thakur, Amitayush
Lee, Jasper
Tsoukalas, George
Sistla, Meghana
Zhao, Matthew
Zetzsche, Stefan
Durrett, Greg
Yue, Yisong
Chaudhuri, Swarat
Machine Learning
Artificial Intelligence
Logic in Computer Science
Programming Languages
Software Engineering
We introduce ${\rm C{\small LEVER}}$, a high-quality, curated benchmark of 161 problems for end-to-end verified code generation in Lean. Each problem consists of (1) the task of generating a specification that matches a held-out ground-truth specification, and (2) the task of generating a Lean implementation that provably satisfies this specification. Unlike prior benchmarks, ${\rm C{\small LEVER}}$ avoids test-case supervision, LLM-generated annotations, and specifications that leak implementation logic or allow vacuous solutions. All outputs are verified post-hoc using Lean's type checker to ensure machine-checkable correctness. We use ${\rm C{\small LEVER}}$ to evaluate several few-shot and agentic approaches based on state-of-the-art language models. These methods all struggle to achieve full verification, establishing it as a challenging frontier benchmark for program synthesis and formal reasoning. Our benchmark can be found on GitHub(https://github.com/trishullab/clever) as well as HuggingFace(https://huggingface.co/datasets/amitayusht/clever). All our evaluation code is also available online(https://github.com/trishullab/clever-prover).
title CLEVER: A Curated Benchmark for Formally Verified Code Generation
topic Machine Learning
Artificial Intelligence
Logic in Computer Science
Programming Languages
Software Engineering
url https://arxiv.org/abs/2505.13938