✨ Data quality: confusion matrix

<Card href="https://humansignal.com/goenterprise"> <img style="pointer-events: none; margin-left: 0px; margin-right: 0px;" src="https://docs.humansignal.com/images/badge.svg" alt="Label Studio Enterprise badge"/> <p style="margin-top: 10px; font-size: 14px;"> This endpoint is not available in Label Studio Community Edition. [Learn more about Label Studio Enterprise](https://humansignal.com/goenterprise) </p> </Card> Returns label confusion matrix with precision, recall, and top confusion pairs. In `ground_truth` mode the matrix is directional: rows are GT labels (actual), columns are annotator labels (predicted). In `all` and `accepted` modes — where no canonical "actual vs predicted" axis exists — the matrix is symmetric. When a task has multiple GT annotations the most recently updated one is used. `top_confusion_pairs.rate` is the share of off-diagonal mass.