300 random Russia-framing documents from C4 (100 per language split), independently annotated by two reviewers using a dual-blind protocol.
Validation dataset for the Rust-based C4 sovereignty classifier (90 signals). Each annotator received only the document ID and a ~500-character text excerpt centered on the matched sovereignty signal. No classifier scores, no automated estimates, no… See the full description on the dataset page:
https://huggingface.co/datasets/CrimeaIsUkraineOrg/crimea-sovereignty-validation.
