A structured rubric for human evaluation of model-generated legal answers and
documents. Each criterion is scored with a small, fixed label set so that
annotations stay consistent and easy to aggregate.
YES
The criterion is fully satisfied.
NO
The criterion fails (at least one violation).
N/A
The criterion does not apply to this response.
Criteria are grouped into five dimensions:… See the full description on the dataset page:
https://huggingface.co/datasets/jusbrasil/indice-jusia.
