# FidelityGPT: Correcting Decompilation Distortions with Retrieval Augmented Generation
> FidelityGPT post-processes existing decompiler output: an LLM labels lines with six distortion types, such as non-conventional dereferencing, redundant variables and confusing control flow, and then rewrites the labeled lines. Its prompts define the distortion types and add retrieval-augmented context from a manually annotated database of distorted IDA Pro and Ghidra lines. A semantic-intensity algorithm chooses which lines to query, and a variable-dependency analysis carries redundant-variable context across 50-line chunks. On 620 function pairs, the authors report 89% detection accuracy and 83% precision, and correction fix rates of 94% (FR) and 64% (CFR) against 83% and 37% for their DeGPT runs.
- Authors: Zhiping Zhou; Xiaohong Li; Ruitao Feng; Yao Zhang; Yuekang Li; Wenbu Feng; Yunqian Wang; Yuqing Li
- Venue: NDSS 2026
- Status: Conference paper
- Publication / posting date: 2026 (exact date not verified)
- Date source: The Internet Society’s Crossref deposit and the official NDSS paper identify publication year 2026. No finer publication date is asserted from the 23–27 February 2026 symposium dates.
- First arXiv submission: October 22, 2025
- arXiv: 2510.19615
- Reviewed arXiv version: v1
- DOI: 10.14722/ndss.2026.230989
- Sources reviewed: 2026-10-18
- Added to the catalog: 2026-10-19
- Original paper:
- PDF:
- Code:
- BibTeX:
- Catalog page:
- Record JSON:
- Chinese version: [简体中文](https://awesomerea.org/zh/papers/paper-fidelitygpt/index.md)
## Scope and limitations
The 620 pairs are mostly -O0 IDA Pro 7.5 functions sampled from Coreutils, curl, ImageMagick, OpenSSL, PuTTY and SQLite builds in a binary-similarity dataset, across ARM-32, MIPS-32 and x86-32, plus x86-64 TheAlgorithms/C code; separate -O1–-O3 and Ghidra sets test generalization. Ground-truth labels and FR/CFR were assessed manually by the author team against the paper’s criteria. Fix rates count lines marked as fixed and are not semantic-equivalence proofs: the authors note that repaired code often cannot be recompiled or executed. Correction runs on whole functions and may degrade on long ones, and the distortion database, built from 150 IDA Pro and 91 Ghidra lines, may miss rarer anomalies. Main results use GPT-4o at temperature 0.5 and remain stochastic; DeepSeek-chat and temperature-0 runs are reported separately. Training-data exposure of the open-source projects cannot be ruled out.
## Why it fits this collection
Adds LLM-based detection and correction of distortions in decompiler output, complementing DeGPT-style refinement and end-to-end LLM decompilation.
## Publication and review notes
The eight-author order follows the NDSS paper page, the PDF, Crossref and arXiv v1, which carries the same DOI. The PDF footer gives ISBN 979-8-9919276-8-0; Crossref lists no pages. The paper’s reference [31] points to the authors’ repository at a 2025-09-07 commit. It has no license file; its datasets contain decompiled functions and source code from third-party projects with their own licenses. The snapshot includes the detection and correction scripts, prompt templates, both distortion databases, seven decompiled-function sets with ground truth and a detection evaluator; the -O1–-O3 and Ghidra evaluation sets were not identified in it. Its README requires manual merging of chunked detection output and states that correction must be evaluated manually using the paper’s Table I. The configuration targets an OpenAI-compatible GPT-4o endpoint at temperature 0 with pinned LangChain 0.1 and Chroma dependencies.
## Primary sources
1.
2.
3.
4.
5.
6.
7.
8.
9.
## About this entry
Reviewed against the paper and its primary sources. Cite the original authors using the linked paper or BibTeX.
Original editorial content: MIT. Linked projects, papers and assets retain their own rights and licenses.