2026 · Preprint
ExeBench-FHD: Addressing Fidelity Issues in Decompiler Evaluation
The paper identifies nine fidelity issues in ExeBench and introduces ExeBench-FHD, adding assembly, data, wrapper and warning fields to improve LLM decompiler evaluation.
Scope and limitations
The evaluation measures GPT 6 Sol's re-execution rate at -O2 on four architectures, including functions that read data sections, mix parameter types or have unused parameters.
Why it fits this collection
It analyzes and improves a benchmark for recovering C code from compiled functions.
Cite
@misc{paper-exebench-fhd,
title = {{ExeBench-FHD: Addressing Fidelity Issues in Decompiler Evaluation}},
author = {{Akarsh Ashok} and {Dinil Mon Divakaran} and {Vivek Balachandran}},
year = {2026},
url = {https://arxiv.org/abs/2610.28212},
howpublished = {arXiv:2610.28212},
note = {Preprint},
eprint = {2610.28212},
archivePrefix = {arXiv},
month = oct
}
Read in a terminal: hf papers read 2610.28212
Publication and review details
- Venue
- arXiv
- Status
- Preprint
- Publication / posting date
- October 20, 2026
- Date source
- First arXiv submission date.
- First arXiv submission
- October 20, 2026
- arXiv
- 2610.28212
- Reviewed arXiv version
- v1
The date and venue come from the arXiv record.
Sources reviewed
Reviewed against the paper and its primary sources. Cite the original authors using the linked paper or BibTeX.