Data for "Can Large Language Models Review the Scientific Literature? A Test Against a 102-Article Benchmark"

Data, prompts, raw tool outputs and figure data for a study testing whether AI tools can carry out the steps of a scientific literature review, against a published, manually constructed review of the Spanish plume literature (102 articles; Schultz et al. 2025, Monthly Weather Review). The study covers four steps: finding the articles, interpreting them, synthesizing across them, and proposing future work. Each experiment folder holds its locked prompt, the original tool outputs, and clean per-article tables. Scripts redraw every figure and recompute the reported numbers. Publisher-hosted article PDFs and the benchmark's full database are not included.