> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/sdgm/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/sdgm/_mcp/server.

# How can I detect data leakage?

> Review temporal data availability and independent evaluation results

Check whether each input value would have been available at the historical prediction time. Common leakage sources include future outcomes in feature columns, backdated events, and historical attributes overwritten with today's values.

Unexpectedly strong evaluation results, a production performance drop, or one column dominating model explanations are reasons to investigate. None proves leakage on its own.

Use [Training Diagnostics](/kumo-engine/user-code-guide/train-a-model/training-diagnostics) and [Model Explainability](/kumo-engine/user-code-guide/train-a-model/model-explainability) to inspect suspicious results. Choose [evaluation splits](/kumo-engine/user-code-guide/train-a-model/predictive-queries#choose-split-assignments) that reflect how the model will be used.