2026 · Preprint
Decaf: Improving Neural Decompilation with Automatic Feedback and Search
Decaf samples multiple candidate decompilations, recompiles them and ranks them using compiler-derived byte distance or a learned reranker. The work studies candidate selection as a way to improve neural decompilation without an iterative source-repair conversation.
Scope and limitations
Byte matching assumes access to the reference compiler/configuration. Experiments subsample ExeBench and exclude examples Hex-Rays cannot decompile; stripped Synth is not execution-scored because call targets cannot be linked. Supplied tests can miss errors. Alternate architectures and obfuscated code are not evaluated, and Juliet results need not generalize to complex vulnerabilities.
Why it fits this collection
Provides a compiler-feedback and candidate-ranking approach with clear distinctions between matching and test-based evidence.
Publication and review details
- Venue
- arXiv
- Status
- Preprint
- Publication / posting date
- May 12, 2026
- Date source
- First arXiv submission date; no separate proceedings or journal publication was verified.
- First arXiv submission
- May 12, 2026
- arXiv
- 2605.11501
- Reviewed arXiv version
- v1
This May 2026 preprint is within the six-month fallback window and is explicitly under review. The pinned code root has no license file; inspected generator/test-set metadata exposes no license declaration.
Sources reviewed
Reviewed against the paper and its primary sources. Cite the original authors using the linked paper or BibTeX.
Primary sources
- https://arxiv.org/abs/2605.11501v1
- https://arxiv.org/html/2605.11501v1
- README.md GitHub · AlexShypula/decaf · Revision f33c8f4
- https://api.github.com/repos/AlexShypula/decaf/git/trees/0d6690fea728fae5d622afa771a7f6b7dd51df44
- https://huggingface.co/api/models/decaf-usenix/Decaf-Gen-22b
- https://huggingface.co/api/datasets/decaf-usenix/Decaf-Test-Sets