TimeSage-MT is a multi-turn benchmark for time series reasoning. Each row in the dataset corresponds to one task in MT_Bench/ and contains:
Task metadata: task id, tier (L1–L4), domain, and the user’s multi-turn questions
Time series payload: the task’s associated time series serialized as a JSON string (time_series_json)
id: task id (e.g. L3_transportation_035)
tier: difficulty tier (L1, L2, L3, L4)
domain: task domain (may be… See the full description on the dataset page:
https://huggingface.co/datasets/Timesage/TimeSage-MT.
