WearableQA

A benchmark for health reasoning over real-world wearable data.

WearableQA comprises 4,084 ten-option multiple-choice questions built from the wearable time series, blood biomarkers, and demographics of 200 real users, each with up to about 500 days of daily measurements. Unlike benchmarks built on synthetic or idealized signals, it preserves authentic wearable distributions — device noise, missing days, and inter-individual variability included.

• 📄 Paper: https://arxiv.org/abs/2609.05405
• 💻 Code: https://github.com/facebookresearch/WearableQA

Quick start

from datasets import load_dataset

ds = load_dataset("facebook/WearableQA", split="test")

ex = ds[0]
ex["question"]   # the complete prompt, ready to send to a model
ex["choices"]    # {"A": ..., ..., "J": ...}
ex["answer"]     # "A"

The dataset is large (median prompt ~86k characters), so streaming is often convenient:

ds = load_dataset("facebook/WearableQA", split="test", streaming=True)

Configurations

The same 4,084 questions in five forms. Ids and answers are identical across all of them — only the way the sensor time series is presented changes.

load_dataset("facebook/WearableQA", "markdown", split="test")     # a different serialization
load_dataset("facebook/WearableQA", "structured", split="test")   # build your own prompts

The representation matters: in our experiments it moved accuracy by several points, and image-based renderings of the same data were far worse than any text form.

Fields — rendered configs (row, col, csv, markdown)

Fields — structured

Everything above except question and representation, plus:

Each record is self-contained — no joins against a separate user table:

ds = load_dataset("facebook/WearableQA", "structured", split="test")
ex = ds[0]
ex["sensor_history"][0]    # {"date": "2023-10-20", "steps": 27858.0, "rhr": 38.0, ...}

# build whatever prompt you want
my_prompt = f"{ex['stem']}\n" + "\n".join(
    f"{d['date']}: steps={d['steps']}, rhr={d['rhr']}" for d in ex["sensor_history"])

Taxonomy

The 16 question types are organized along two complementary axes:

• Data vs. health reasoning — computing over longitudinal measurements (correlations, excursion counts, recovery times, trend shapes) versus interpreting them physiologically (risk assessment, differential diagnosis, prognostic prediction).
• Single- vs. cross-signal reasoning — reasoning within one metric versus integrating several.

Ground-truth answers are balanced uniformly across options A–J within each reasoning group, so the random baseline is 10%.

Citation

@misc{lee2026wearableqa,
      title={{WearableQA}: A Benchmark for Health Reasoning over Real-World Wearable Data},
      author={Ji Soo Lee and Xilun Chen and Pierce Chuang and Ashish Shenoy and Jason Wei and Dohwan Ko and Hyunwoo J. Kim and Benoit Corda},
      year={2026},
      eprint={2609.05405},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2609.05405},
}

License

The data is licensed under Creative Commons Attribution-Non Commercial 4.0 International (CC BY-NC 4.0), and subject to the following additional terms: (i) No re-identification or attempted re-identification; (ii) No use in connection with clinical, diagnostic, or treatment decisions; (iii) No use in a manner that is discriminatory, harmful, or misleading with respect to health-related outcomes; (iv) The Dataset is provided "as is", without warranties of any kind, whether express or implied, including without limitation accuracy, completeness, or fitness for a particular purpose, and is provided for research and benchmarking purposes only.