- entity occurrences table to html, add caption
- entity types table to list
- extraction input and commit table to html, add caption
- completeness failure log table to html, add caption
- supporting files table to html, add caption
- dataset structure subsets table to html, add caption, split field names to single rows for legibility
- original data producers table to html, add caption
- glossary table to list
Rows are stored in document order and the two extraction passes are merged rather than appended. Previously this was one sentence in Subsets and absent from Processing Steps.
NER_v1: merge the retry pass and store rows in document order
Adds 188,957 rows from re-running the windows that returned no answer, for 24,035,075 in total. Both passes are merged and written in record, page, window order rather than appended, so the retry rows are interleaved. Sorting colocates rows from the same record, which reduces the Parquet set from 1,372 MB to 939 MB with identical content.
Document that authentication is required while the dataset is private
The loading examples omitted a token. Because the repository is private, the Hub reports it as not found and `datasets` falls back to the local cache instead of raising an authentication error, which surfaces as a misleading message about a missing config. Examples now pass token=True and the cause is described.
