# Changelog

## 2.0.0 - 2026-07-19

- Replaced executable dataset loading with script-free Parquet configurations.
- Made `all/full` the default unsplit corpus with 2,710,740 non-empty rows.
- Added `source_id`, `label_name`, `is_label_conflict`, and `label_variants` fields.
- Added a deterministic 10,000-row `sample/full` configuration balanced across language and label.
- Added reproducible build and validation scripts with checksummed reports.
- Retained the original gzip files for archival reproducibility without exposing their row-level splits as Hub configurations.

This release intentionally defines no train, validation, or test split.

The archived revision is available at <https://doi.org/10.57967/hf/9639>.

## 1.0.0 - 2023

- Published the original Many Emotions dataset.

The archived revision is available at <https://doi.org/10.57967/hf/9638>.