kambabench/docs/DATASETS.md

# Datasets

**Status: not yet written.** Roadmap item v0.1 ("Ingest candidate datasets" / "Establish
provenance records").

This document will hold, for each candidate source corpus (DDD-Kenya `Kamba-ASR-Data-Subset-484H`,
Thiomi, Google FLEURS `kam_ke`, Meta Omnilingual ASR):

- exact license and any redistribution restrictions,
- full provenance (who collected it, how, and when),
- speaker composition (count, gender/age distribution where available),
- recording conditions (studio, field, phone-recorded, etc.),
- transcription methodology (read-aloud, translate-then-record, natural-speech-then-transcribe),
- domain/theme coverage, from `domain_audit.py`'s output,
- dialect coverage, and
- relationship to the other candidate sources, from `dedup_check.py`'s output.

See the root `README.md`'s "Candidate source corpora" table for the current high-level summary
pending this document.