An open workbench for sharing how Kaggle research actually gets made.

Start here: 🦆 Take a seat at the OpenKaggle bench · Contributing · Publishing · Citation · Support

OpenKaggle is a small, open workbench for sharing the real texture of Kaggle research: code to try, experiments to compare, notes to learn from, and ideas that are still taking shape.

It is a place for competitors, students, engineers, and curious builders who like making things in public. Bring a notebook, a useful script, a clean baseline, a surprising result, a careful correction, or a question.

We care about where work came from and what was actually tested, but you do not need a polished paper to join.

Bring a notebook, a question, a useful correction, or just curiosity. One sentence is enough to start; membership is optional. Visit the pinned OpenKaggle bench issue and write:

am I at the bench? 🦆

The bench will send a membership invitation when capacity is available.

OpenKaggle is a place to share:

• current and archived Kaggle competition projects;
• exploratory notebooks and reusable analysis systems;
• data preparation, training, inference, evaluation, and submission processes;
• technical production workflows that make research more reliable;
• experiment logs, including negative and inconclusive results;
• research plans, hypotheses, decision records, and open questions;
• provenance records, checksums, environment details, and reproduction receipts;
• open datasets and models when their licenses permit redistribution.

Each repository should plainly say what it is for, what someone can reuse, what evidence supports its claims, and what remains outside its scope.

There are several good ways to participate:

• request organization membership by leaving the short comment in the pinned OpenKaggle bench;
• propose a repository for the OpenKaggle organization;
• keep a repository under your own account and ask us to list it in the portfolio;
• contribute a focused improvement to an existing project;
• reproduce an experiment and report what matched or differed;
• publish an experiment log, research plan, analysis system, or technical workflow;
• help clarify documentation, provenance, licensing, or known limitations.

A small, well-explained contribution is often more useful than a large unexplained upload.

This is a growing map of shared tools, careful archives, and competition work.

OpenKaggle does not ask a project to fit a fixed maturity ladder. A README should simply state the question, inputs, method, evidence obtained, reproduction path, and known limits. If a repository is an archive or is actively maintained, say so in plain language.

Local validation, public leaderboard scores, private leaderboard scores, and estimates are different kinds of evidence. Repositories should not blur them together. Failed experiments belong here too: when the setup and outcome are recorded, failure becomes reusable knowledge.

We preserve as much research context as possible without pretending that every public file may be redistributed.

• Each repository states its own license and status.
• A code license does not automatically cover datasets, model weights, competition files, or third-party outputs.
• Original licenses and competition rules take precedence.
• Large permitted artifacts may live in releases, Git LFS, Kaggle datasets, or another documented store.
• When bytes cannot be mirrored, we prefer a source URL, version, checksum, expected path, approximate size, and retrieval instructions.
• Credentials, private identifiers, restricted data, and unrelated personal files do not belong in public commits.

The practical release gate is documented in Publishing research from a competition workspace.

For a consistent way to cite a repository, model, report, or artifact release, see the OpenKaggle citation and archival standard. It uses CITATION.cff, a synchronized BibTeX entry, a README citation block, and a versioned release or DOI when one exists.

Unclear cases are discussed patiently. It is always acceptable to publish the method and provenance record before publishing the artifact itself.

OpenKaggle is small on purpose. Some projects arrive tidy; others are still on the workbench. That is fine. A useful notebook, a negative result, a clean reproduction, or one well-described question can all be worth sharing.

We care about credit, provenance, and boundaries, while leaving room for curiosity, humor, and unfinished work. If you find something useful, cite it, try it, tell us what changed, or bring your own version. There is room at the bench.

OpenKaggle is an independent project and is not affiliated with, endorsed by, or operated by Kaggle or Google. Kaggle and related names and marks belong to their respective owners.