2026 · Preprint
From Dead Code and Static Requirements to Working Engines: Software Revival with Coding Agents
ReviveBench evaluates repair of nonrunning software and source-unavailable implementation of bounded computational and transactional engines from open specifications, using hidden reference-based checks. It also audits grading defects that can wrongly accept or reject reconstructions.
Scope and limitations
Studies ten revival and thirteen bounded engine tasks; reconstruction receives public specifications and three worked examples. Most model/task cells have one or two seeds; the reconstruction matrix reports the best eligible run. Excluded infrastructure failures and differing API adapters limit comparisons. Task-level pass thresholds do not require every check to pass, and CFD passes cannot establish numerical-solver capability. Verifier audits and contamination controls do not prove general equivalence or rule out prior exposure.
Why it fits this collection
Connects software reconstruction with audits of the tests used to judge its correctness.
Publication and review details
- Venue
- arXiv
- Status
- Preprint
- Publication / posting date
- September 28, 2026
- Date source
- First arXiv v1 submission; no formal venue publication was verified.
- First arXiv submission
- September 28, 2026
- arXiv
- 2609.36161
- Reviewed arXiv version
- v1
- DOI
- 10.48550/arxiv.2609.36161
The four-author v1 identity matches arXiv and DataCite, which registered the arXiv DOI on 2026-09-30. The article uses the arXiv non-exclusive license. No author-linked public benchmark/verifier package or its code/data license was verified in the reviewed full text and focused search. No software packages, model weights, benchmark tasks or experiments were downloaded or executed.
Sources reviewed
Source review only; experiments were not rerun for this collection. Cite the original authors using the linked paper or BibTeX.