BCINexus · Publishing
Publishing datasets
Datasets go through the same publication state machine as studies — not a lighter parallel path. What differs is what counts as complete.
The two hard requirements
| Requirement | Why it blocks |
|---|---|
| At least one uploaded version | A dataset with no version has nothing to publish. |
| An attached paradigm | Without the NeuralFlow definition of the task and the labels, a recording cannot be understood or reused. It is the difference between a dataset and a pile of numbers. |
Title, signal modality, task type, and a summary of at least 30 characters are required as well, exactly as for studies. Everything else — licence, keywords, byline, affiliations, references, ethics and consent disclosure — is a warning that drives the readiness score.
Which checks are skipped
Results and reproducibility checks do not apply to datasets. A raw upload was never meant to report training metrics, so it is not warned about missing them. Details are in publish readiness.
Publishing one
- 1
Create the dataset and upload a version
Versions are immutable snapshots. Adding data later means adding a version, not editing the old one.
- 2
Attach the paradigm
Export the NeuralFlow paradigm that defines the task, trials, and labels, and attach it to the dataset.
- 3
Describe provenance honestly
Where the recordings came from, how many subjects and sessions, the acquisition hardware, and any preprocessing already applied. This is the section reusers read first and the one reviewers question most.
- 4
Set a licence and disclose ethics
For human-subject modalities, state ethics approval and whether informed consent was obtained — or state explicitly that neither was required.
- 5
Submit and be reviewed
Datasets enter the same moderation and review path as studies, with the same states and the same audited transitions.
Access and roles
- Reader
- View the dataset and download its versions.
- Editor
- Everything a reader can do, plus submit new versions.
- Public
- Once published, anyone can find and download it under its licence. Free accounts have a weekly community download allowance.
Integrity and reuse
- SHA-256 integrity hashing is available on the Lab plan and above, so a reuser can confirm they have the same bytes you published.
- Datasets can be linked to the studies that use them, which is what populates the provenance edges in the knowledge graph.
- Because versions are immutable, a study citing version 2 keeps meaning what it meant after version 3 lands.
Privacy before you upload
De-identify first
Publishing is a one-way door in practice, because other people begin downloading and citing immediately. Strip participant identifiers, free-text notes, and session metadata that could re-identify someone, and confirm your consent forms actually permit public release. If in doubt, share to a private team first — see teams and labs.
Related
Publish readiness
The exact checks, including the dataset-only ones.
Publishing a study
The shared state machine.
Analysis Suite
Preparing recordings before they are worth publishing.
Browse datasets
What is already published.
Dataset objects and their versions are addressed by the same identifiers throughout the app and the API — see /docs/bcinexus/api.
Something missing or out of date? Email support or request a doc page. Include the page name and what you expected to find.