Skip to main content
A submission closes one comparison; it does not end the workflow. Record what the result supports, create a new experiment, and test the next hypothesis on the appropriate public dataset. When source release is enabled, you may study the source explicitly released by the competition. Treat it as a control or a source of transferable patterns, not permission to use private material or hardcode answers. Return to compare and record analysis to preserve the reasoning, then repeat the evaluation loop with a clear baseline.