Software Engineering Empirical Research Radar

Assessing the Impact of Code Changes on the Fault Localizability of Large Language Models

Paper detail page in SEER Radar.

Authors

Sabaat Haroon, Ahmand Faraz Khan, Ahmad Humayun, Waris Gill, Abdul Haddi Amjad, Ali R. Butt, Mohammad Taha Khan, Muhammad Ali Gulzar

Venue / Year

ICST 2026

Topics

Debugging / Fault Localization / Diagnosis / Repair; AI / LLM for SE; Testing / QA

Abstract / Summary

A large-scale robustness study evaluates ten LLMs on 750,013 fault-localization tasks from more than 1,300 Java and Python programs. Semantic-preserving changes cause models to lose previously correct localizations in 78% of examined cases, exposing sensitivity to syntax and context order.

External Links

Official / Source Page

PDF

Local PDF is not available on SEER Radar yet. When a public source is recorded, this page will add a local reading link with source attribution.