ScholarCatalyst Benchmark for Retrieval of Research Inspiration
October 2, 2026
ScholarCatalyst evaluates the ability to retrieve papers that drive research progress based on author-labeled rationales. In retrieval tasks using research questions, agentic search methods achieved a Recall@20 of 0.42, underperforming standard embedding retrieval at 0.48.
HOW THIS AFFECTS YOU
●
researcherYou can use this benchmark to evaluate how well LLM agents navigate scientific literature for idea generation.