---
pretty_name: WildChat Annotation Pool
license: other
task_categories:
  - text-classification
language:
  - en
tags:
  - safety
  - annotation
  - conversations
size_categories:
  - 10K<n<100K
---

# WildChat Annotation Pool

This dataset is a filtered English-only annotation pool derived from `allenai/WildChat-4.8M`.

## Contents

- `conversations.jsonl`: conversation records to annotate
- `pool_manifest.jsonl`: lighter-weight manifest with stable ordering metadata
- `metadata.json`: filter counts and generation summary
- `subset_spec.yaml`: filtering spec used to build the pool
- `sample_review.jsonl`: small manual QA sample from the final pool

## Filtering Summary

- source dataset: `allenai/WildChat-4.8M`
- rows scanned: `100001`
- rows kept: `19228`
- language: English only
- additional filters: redacted removal, turn-count and length bounds, empty-turn rejection, exact duplicate removal, simhash near-duplicate removal

## Notes

This is intended as a frozen pool for human annotation workflows. It is not itself a label dataset.