MoodTrace-20D is a longitudinal emotion dialogue benchmark scaffold derived from daily spoken English conversations (transcribed) over ~20 consecutive days, organized as pair-level JSONL records. This HuggingFace repository currently provides:

Dataset card + basic dataset statistics (stats.json) Links to the benchmark loader/evaluator code

Raw conversation text is not released in this repo at this stage (privacy-first release).

Benchmark code

GitHub:… See the full description on the dataset page: https://huggingface.co/datasets/Yukyin/moodtrace-20d.