How to contribute

The platform grows through its community: clinicians contribute anonymized headache cases, and researchers contribute benchmark results produced on the published dataset. This page collects the guidelines for both (FR-12).

Who can contribute

Contributing requires a registered account with your professional context. An administrator vets every account before submission functions unlock; case submission additionally requires a recorded center affiliation. Browsing cases and downloading the dataset stay open to everyone, no account needed.

Case structure

A case submission contains exactly the fields of the study protocol's case schema — nothing more:

  • Age in years and sex
  • Structured anamnesis covering symptomatology, duration, accompanying symptoms, and triggers
  • Clinical/neurological findings (optional)
  • ICHD-3 classification: one validated primary diagnosis, plus further validated diagnoses if applicable
  • Origin: whether the case describes a real (anonymized) patient or is synthetic
  • Language the case is written in

The form deliberately contains no fields for excluded identifiers — there is no place to enter them.

You can submit cases one at a time through the form, or upload many at once as a CSV/Excel batch — both use the same schema and the same review. The batch is the public-dataset counterpart; study-tier cases use the separate encrypted study upload instead.

Anonymization rules

Cases must be irreversibly anonymized before submission. In particular, a case must contain none of the protocol's excluded identifiers:

  • no name
  • no date of birth
  • no address components
  • no case number
  • no presentation date
  • no treating persons

Beyond this list, leave out any rare, potentially identifying detail (an unusual combination of profession, locality, and circumstances can identify a person even without a name). You confirm this checklist with every submission, and reviewers check it independently.

Validation protocol

Every submitted case is reviewed by a headache-specialist reviewer before publication. The review checks two things: the ICHD-3 classification is clinically valid for the case, and the anonymization is sufficient — no identifiers or re-identification risks remain. A case is published only when both checks pass; otherwise it is rejected with feedback. Review decisions are recorded immutably.

You follow the status on your dashboard, which also carries the reviewer feedback. Notification e-mails contain status information only — case content never transits e-mail.

Contributor terms

By submitting a case you confirm that you hold the rights to contribute it and accept that accepted cases are published under the platform's dataset terms, with your name shown as contributor on the case and on the contributors page. If you later delete your account, accepted cases remain in the dataset with your attribution removed — published dataset releases cannot be rewritten. For real cases you additionally attest that collection and anonymization followed the approved study protocol.

Submitting benchmark results

The platform hosts externally produced benchmark results; it does not run evaluations itself. It only considers datasets and benchmarks that rest on ICHD-3 diagnoses: the cases carry ICHD-3 reference diagnoses and models are scored against them — at diagnosis level (the exact ICHD-3 code) and at group level (the ICHD-3 diagnostic group, e.g. 1. Migraine). Registered researchers submit results via Submit a result. The methodology fields are mandatory:

  • Model name and version/date
  • Prompting strategy
  • The dataset release the result was produced on
  • Top-1 and Top-3 accuracy at diagnosis level, Top-1 accuracy at group level (Top-3 at group level and Cohen's κ optional)

Reproducibility links (code repository, paper/preprint) and cost metadata are optional but strongly encouraged — entries that others can reproduce carry more weight.

Submitted results are moderated before publication and appear with the badge unverified (self-reported). The verified badge is reserved for results the platform team has evaluated or independently reproduced. Results taken from a peer-reviewed journal article or proceedings paper carry the separate peer-reviewed publication badge: peer review checked the publication, which is not the same as the team reproducing the numbers.

Benchmark requests

Instead of self-reporting, you can commission a benchmark run by the platform team via Request a benchmark: name the target dataset release, the model, the method and prompting strategy, an access path (code repository, container, or API access), and a contact. The team evaluates manually and off-platform, ideally on the hold-out subset. Your dashboard tracks the request status (received → in evaluation → published / declined); a published run enters the leaderboard as a verified entry.

Start here