Hugging Face has introduced Spark-to-Paper, an end-to-end research paper generation system that operates as a set of 13 composable skills within existing coding assistants, eliminating the need for separate agent platforms or orchestration services. The system is designed to transform a one-line research idea into a complete manuscript with verified citations, auto-run experiments, and editable vector figures.
The core innovation lies in separating model-based judgment from deterministic operations, and experiment planning from reporting. This ensures that required evidence is specified before results are observed, and manuscript claims are revised based on measured outcomes. To enhance reliability over long research trajectories, the system integrates deterministic integrity checks with self-critique, addressing a failure mode termed the 'Self-Refutation Loop' where repeated experiments reject the original research objective.
In controlled evaluations across eight research topics, Spark-to-Paper achieved 99.5% citation validity and 96.4% figure editability. A controlled ablation study showed that fabrication detection improved from 14% for a single-pass draft to 92% with the full integrity and review stack, while adversarial review reached 74% precision. The full system consumes 11.9M tokens, costs $8.1 per manuscript, and takes an average of 3.2 hours to complete.
These results demonstrate that end-to-end research paper generation can be implemented as a lightweight, composable workflow inside existing coding assistants, keeping experimental evidence central to how claims are accepted, revised, or abandoned. Spark-to-Paper is now featured on Hugging Face Daily Papers, allowing users to drop in a one-line research idea and receive a complete draft.