Submit your results

Submissions are made as GitHub issues on the RelBench repository: you fill a short form (method name, links, an in-context checkbox) and attach your prediction tables — one CSV per task, named <dataset>__<task>.csv. The submission is validated automatically against the official test tables; a leaderboard (classification / regression / recommendation) is scored only when every task in it is present and valid. A maintainer reviews the validation report and, on approval, your entry is published. See the submission guide.

1. Validate and package locally. This scores every prediction table against the test sets, prints the per-leaderboard verdicts, and writes a clean submission zip.

   python -m relbench.leaderboard <submission_dir> --package

2. Open a submission issue. Fill in the form and drag the zip into the attachment area.
   Open a submission issue ↗
3. Automatic validation, then review. The validation report is posted as a comment on your issue within minutes. Once a maintainer approves, your entry appears on the leaderboard.