Swiss rulings, same rules, own files
The two Federal Supreme Court sets are being curated now, beside the other pack. They are written to their own files, not mixed into that pack.
Legal area is through this pass. The note is the facts. The court's considerations are not stored. The label is the dataset's own sub-area, 13 of them. Train kept 9,854 of 10,475. Rows that did not fit were dropped, not cut. Validation kept 2,872 and test kept 7,577.
Judgment, dismissed or approved, is downloading now. German, French, and Italian are the original rulings. The English file is only the machine translation, and only that file has to fit the 2,048-token window used by the other English notes. Original-language facts may run to 24,000 characters, because that file is for Apertus.
A row is dropped when the facts are empty or when a court-reasoning heading is in the text. The English translation is capped at 40,000 train rows. Both sets are CC BY-SA 4.0.