README.md

---
license: mit
language:
- en
size_categories:
- n<1K
task_categories:
- text-classification
configs:
- config_name: default
  data_files: data/train-*.parquet
---

# CompOrca-gold

> **Paper:** [CompOrca: Corpus-Scale Compliance Labelling of Instruction-Tuning Data](https://arxiv.org/abs/2609.37807)  
> Philipp E. Glass and Alina Miron. PlurVA-LLM Workshop @ AACL-IJCNLP 2026.

450 rows of [`cemiu/CompOrca`](https://huggingface.co/datasets/cemiu/CompOrca)
labelled COMPLIANCE or NONCOMPLIANCE by two human annotators.

Source: [`Open-Orca/OpenOrca`](https://huggingface.co/datasets/Open-Orca/OpenOrca),
split `train`.

Columns: `idx`, `id`, `system_prompt`, `question`, `response`,
`annotator_1` (all 450 rows), `annotator_2` (150 rows, null elsewhere), and
`judge`, the label from
[`cemiu/CompOrca`](https://huggingface.co/datasets/cemiu/CompOrca)
(null on the 150 rows where the five passes disagreed).

MIT, as with the underlying OpenOrca text.

## Citation

```bibtex
@misc{glass2026comporca,
  title         = {{CompOrca}: Corpus-Scale Compliance Labelling of Instruction-Tuning Data},
  author        = {Glass, Philipp E. and Miron, Alina},
  year          = {2026},
  eprint        = {2609.37807},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CL},
  url           = {https://arxiv.org/abs/2609.37807}
}
```