saxenan3/welfare-qa

WelfareQA — reading copy

Editorial reader for the **WelfareQA** dataset.

A cross-country welfare-AI evaluation dataset for India, Nigeria, and Kenya. 1,500 ground-truth canonical rows on a shared 71-column schema, expanded to 9,385 multilingual variants. Verified to April 2026 against named government sources.

This Space is a static, editorial reading copy: cover plate, three frontier-model failure cases, the gap proof, and a sample of canonical rows browsable by country, section, and language.

For the full dataset and schema, see the dataset card.

Built with Adaption. Submitted to the Uncharted Data Challenge.