Lacuna: A Research Map for Machine Learning

A new machine-learning research map called Lacuna uses large language models to convert academic papers and metadata into structured summaries, concept elements, and research proposals, according to a preprint posted to arXiv on June 24, 2026 [1]. The system transforms scholarly records into markdown summaries while preserving links to primary source records and supporting papers. Interfaces are provided through web, markdown, and MCP channels [1]. The map is designed to help researchers navigate the growing volume of machine-learning literature, a challenge that has intensified as preprint servers and open-access journals accelerate the pace of publication [7]. On the LitSearch retrieval benchmark, Lacuna achieved a Recall@10 of 0.538, compared with 0.424 for OpenScholar v3 [1]. The authors also tested the system on Multi-XScience-CS/ML and ScholarQA-CS-ML, where it outperformed OpenScholar across all three datasets [1]. The team further evaluated Lacuna Deep Research, a multi-stage report agent that operates over the map, on 25 ReportBench-ML survey tasks. Lacuna Deep Research recorded a citation F1 of 0.052, citation precision of 0.339, 99 expert-reference hits, and a RACE report quality score of 7.82 out of 10. By comparison, GPT-Researcher posted an F1 of 0.039, precision of 0.290, 72 hits, and a RACE score of 5.24 out of 10 [1]. The work arrives as the open-science movement continues to push for greater transparency and accessibility in research dissemination, with practices such as open-access publishing and data sharing gaining traction across STEM disciplines [7]. Tools that structure and summarize scholarly output align with that broader effort, though the authors note that each generated item retains provenance links to its source records [1]. The preprint was submitted to arXiv’s Digital Libraries section and is hosted under the arXivLabs framework, which allows community collaborators to develop and share new features on the platform [1].