> 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 does Kumo handle cold start?

> Evaluate predictions for entities with limited history

Kumo Engine can use entity attributes and related records when an entity has little interaction history. Prediction quality depends on those available signals and the selected model.

Evaluate new and low-history entities separately. For recommendations, check whether the model supports scoring candidates added after fitting. A newly introduced item and an item absent from the training graph are different cases.

The [New Arrival Recommendations example](/kumo-engine/get-started-with-kumo-engine/workflows-and-solutions/personalization/new-arrival-recommendations) ranks recently introduced products that were present during fitting and have recorded purchases. It does not demonstrate scoring unseen candidates.