RSI4Science
English

Contribute a dataset

RSI research needs scientific tasks with checkable outcomes, feedback for revisions and independent comparisons. Existing measurements, simulations and experimental records can be starting points. Bring a prepared package or start with a source link and example.

A paper is optional. Share original data or curate an existing resource with clear sources and reuse terms.

Keep your data where it is

Share a versioned link from Zenodo, Hugging Face, GitHub or an institutional repository, with access terms and a representative public example. The full dataset can stay with its current host. How linked datasets are reviewed.

Start with an email. Draft preparation stays on your device; email the project lead to discuss a contribution.

Dataset
Scientific records with documented sources, fields and reuse terms. Unlabeled data can begin as a source suggestion.
Task
A question with defined model inputs, expected outputs and a way to check them.
Benchmark
A collection of tasks and a fixed procedure for comparing performance.
For a checkable task package

Follow the package guide for the required files, then run local checks before requesting human review.

The current checker accepts JSON records with scalar fields and reference targets; other formats begin as a source suggestion.

Package files and examples

Suggest a dataset or source

A few sentences and public links are enough to begin. This form creates a draft on your device; it does not upload or submit it.

Fields are required unless marked optional. A few sentences are enough; write “Not yet known” where needed. Use public links, without private data or protected assessment answers.

You can start either choice with public links and a short description. An RSI study is optional. Drafts stay separate when you switch between these choices.

Describe what was measured, computed or generated. For one possible scientific task, what would a model receive and what would it predict, choose or produce? Mark any unknowns.

Share a versioned dataset URL and the original paper, if available. Include the license or access terms, file format, approximate size and your role. The full dataset can stay with its current host; do not paste the data itself.

Link a public example with its fields and units. Describe any reference or checker, feedback a researcher could use after a prediction, and records from the same family or experiment. State what is still missing.

Contributors and further details (optional)

List everyone who worked on this resource, one person per line: name and contribution. Institution and city are optional.

A team shares one dataset listing. Map details are checked with the team before publication; paper affiliations can be linked to their source.

If known, describe how a researcher could use this resource to test an improvement. You can leave this blank.

Creates a preview below. Save or copy the draft to keep it on your device. This page does not upload your responses.

Ready to discuss it? Save or copy your draft, then email the project lead and attach it if you wish. Your email application opens; you choose what to send. How drafts are handled (English).

Switching language temporarily keeps all drafts in this tab to restore them on the next page. This temporary copy is removed when restored; it expires after two minutes.

How a dataset joins the collection

Source listings receive metadata, access and relevance review. Checkable packages also receive data, scoring and reproducibility review. Each listing states what was examined and which uses it supports.

Package files and examples · Review criteria

Sources and permissions

Use public source links in your draft. Each resource retains its own access and reuse terms. We confirm permission before redistributing data; private drafts and protected evaluation answers stay private. Sources and reuse notes.