Kurate: Scalable Scientific Quality Analysis

Scientific search systems can retrieve relevant papers without assessing the quality of the associated evidence. We present Kurate, an AI-assisted system for converting publications and associated scientific artifacts into structured, source-traceable evidence-quality assessments. In a snapshot of 2,448 processed papers (incl. 2399 randomized trials), Kurate assessed 8 methodological and reporting dimensions: statistical power, causal identification, preregistration, selective reporting, measurement validity, analysis prespecification, reporting transparency, and conflict of interest. Across the corpus, the most frequent weak signals were selective reporting, statistical power, and analysis prespecification, while domain-level summaries showed different quality profiles across clinical areas. We also examine the SPRINT blood-pressure trial as a worked example, showing how a paper-level grade decomposes into judgments linked to registration, protocol, participant-flow, and scoring evidence. This evaluation demonstrates the potential for Kurate to provide scalable evidence appraisal for meta-scientific research questions.