---
pretty_name: MetroTalk Core
language:
- en
task_categories:
- text-generation
---

# MetroTalk Core

MetroTalk Core contains normalized public-transit support dialogues for response-quality research.

## Processing

Records retain the original turn order while standardizing speaker labels and whitespace.

## Intended use

Use this card for dialogue-response evaluation and controlled prompt experiments.

## Verified upstream provenance

- **Verified upstream dataset:** [`toolathlonEval/MetroTalk-Archive-2023`](https://huggingface.co/datasets/toolathlonEval/MetroTalk-Archive-2023)
- **Release marker:** `metro-dialogue-2023-11`
- **Evidence:** A catalog search for the upstream dataset name returned several similarly named candidates, so discovery alone was ambiguous. Provenance was instead verified using only the repository-linked raw evidence documents recorded in `provenance/source-map.md` of the GitHub repository `toolathlon-jd/CivicText-Provenance-Desk`: the linked release note and the linked checksum record independently agree on the upstream dataset identity and the release marker.