Datasets

The public dataset is openly downloadable. Two kinds of private data are shown as aggregate figures only: a held-out synthetic test set for contamination-resistant evaluation, and the study dataset, held only as encrypted blobs.

Public dataset

Current release
1.1
Published
2026-10-22
Cases
912 case variants
License
CC BY-NC 4.0
How to cite
headache.bench contributors. Community headache case dataset, release 1.1 (2026-10-22). Includes material adapted from: Zwanzig D. Primary Headache Vignettes LLM Generated (Oct. 2026): synthetic ICHD-3 headache vignettes, chapters 1–4. headache.bench. | de Andrade JBC, da Silva Costa TB, Vasconcelos JL, Lopes TLM, Balsells MD, Cristofolini VL, Querobin SO, Filho FMR. Validation of an AI-based platform for structured diagnosis of headache disorders using ICHD-3 criteria. BMC Neurology 25, 489 (2025). doi:10.1186/s12883-025-04473-1. Dataset: Andrade J. Head.AI Platform. Mendeley Data, V2 (2025). doi:10.17632/4fms6cw2s5.2

Downloads & release history

Held-out synthetic test set

Synthetic cases the platform team keeps back from every public release, so models can be evaluated on cases they cannot have seen. Not downloadable.

Version
1.0
Cases
364 texts, 70 ICHD-3 diagnoses

Aggregate figures

Private (study) dataset

Cases collected under ethics protocol BB 211/25, held only as encrypted, never-decryptable blobs — not publicly downloadable. The figures are study-team-maintained aggregates.

No study-dataset figures have been published yet.

Aggregate figures