Verified up to K: A Stated Verification Bound Anchors LLM Repairs of Buffer Overflows

This pre-registered study tested whether stating a verification bound changes the buffer sizes selected by three language-model coding systems. Across 2,862 repair attempts on 53 C and C++ contest programs, increasing the stated bound shifted the selected buffer sizes in all three systems. With the bound set to the size of the failing input, the full-suite pass rate fell from 80% to 32% for Claude and from 96% to 76% for Codex. On six C units amenable to a separate ESBMC check, all 15 examined repairs that failed the broader tests were certified at the stated bound. The paper examines how providing the actual input constraint or explaining the scope of the verifier affects repair outcomes.

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-19
DOI
https://doi.org/10.5281/zenodo.24374714
Primary Topic
Software Testing and Debugging Techniques
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Verified up to K: A Stated Verification Bound Anchors LLM Repairs of Buffer Overflows

S. M. Abdulla Al Mamun
Zenodo (CERN European Organization for Nuclear Research)
Software Testing and Debugging Techniques
preprint

Verified up to K: A Stated Verification Bound Anchors LLM Repairs of Buffer Overflows

S. M. Abdulla Al Mamun
preprint en

Abstract

This pre-registered study tested whether stating a verification bound changes the buffer sizes selected by three language-model coding systems. Across 2,862 repair attempts on 53 C and C++ contest programs, increasing the stated bound shifted the selected buffer sizes in all three systems. With the bound set to the size of the failing input, the full-suite pass rate fell from 80% to 32% for Claude and from 96% to 76% for Codex. On six C units amenable to a separate ESBMC check, all 15 examined repairs that failed the broader tests were certified at the stated bound. The paper examines how providing the actual input constraint or explaining the scope of the verifier affects repair outcomes.

Zenodo (CERN European Organization for Nuclear Research)
University of Manchester (GB)
Software Testing and Debugging Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon