How confident should you be about cluster assignments?

Clustering algorithms assign data points to groups without saying how confident those assignments are. This work wraps clustering in a statistical framework that produces valid confidence sets for cluster labels—basically, ranges you can trust. The trick: cluster labels aren't true labels, so standard confidence methods fail. They developed a weighted conformal approach that corrects this mismatch, with proven coverage guarantees. Experiments show tighter, more informative confidence sets than existing methods, especially in high-dimensional data.