2026 · Preprint
SWE-Game: Can Coding Agents Build the Games We Want?
Evaluates construction, completion, repair and Godot-to-Unity porting against executable reference games. Evaluator-owned probes and input replays check gameplay behavior, while visual rubrics assess presentation separately.
Scope and limitations
The 247 tasks derive from 41 games. Inputs vary by task; porting receives Godot source, so this is not purely black-box reconstruction. Each model uses one framework, confounding their effects. Finite scenario checks and 100-game functional validation do not establish general gameplay equivalence.
Why it fits this collection
Adds executable checks for reference-guided game reconstruction and cross-engine behavior preservation.
Publication and review details
- Venue
- arXiv
- Status
- Preprint
- Publication / posting date
- September 27, 2026
- Date source
- First arXiv submission; the reviewed version is identified separately.
- First arXiv submission
- September 27, 2026
- arXiv
- 2609.33678
- Reviewed arXiv version
- v3
- DOI
- 10.48550/arxiv.2609.33678
First submitted September 27; reviewed v3 dated October 7. No author-linked code or dataset release was verified in the reviewed manuscript.
Sources reviewed
Reviewed against the paper and its primary sources. Cite the original authors using the linked paper or BibTeX.