Curated SDRF files currently come from the BigBio GitHub site (https://github.com/bigbio/sdrf-annotated-datasets) where a group of community curators, led by the developers of SDRF-Proteomics itself, produce and deposit high-quality SDRFs for previously deposited datasets. The curators may not understand the datasets as fully as the original authors, but they understand best how to create valid SDRF files.
Repository SDRF files are the original files deposited into a ProteomeXchange repository by the original data submitters along with the original data files. The submitters presumably know their data the best, but often are less familiar with producing valid and high-quality SDRF files. Many repository SDRFs do not pass validation, but may still be useful for understanding the dataset.
Agentic SDRF files are produced by the HAMLET pipeline (https://github.com/NCEMS/HAMLET), which aims to analyze large numbers of PXDs and the associated journal articles with generative AI agents in order to produce SDRF files for many more datasets than is possible by human curators. These SDRFs are almost always technically valid, but some values may not be accurately captured by the AI. This pipeline is still in a beta phase and these SDRFs should be evaluated carefully for possible incorrect values.
