Evidence-backed facts
- OpenML documents machine-learning datasets, tasks, flows, runs, evaluations, studies, files, and metadata. informative
- Public reads are separate from authenticated uploads and edits; access does not grant dataset, task, run, model, benchmark, or OpenML authority. informative
- Clients must bind resource type and ID, dataset version, task, flow, run, evaluation measure, file, API key, and pagination. informative
- Contributor-supplied data and results can be incomplete, incompatible, duplicated, biased, or invalid and do not prove reproducibility, fairness, safety, or benchmark superiority. informative