Cluster Map

Regions projected onto their first two principal components. Identity clusters (left) are derived from D-scores; terroir clusters (right) from Layer 2 TF-IDF. If terroir determined identity, the two plots would show the same structure. They do not.

Identity Clusters
D-scores → PCA → K-means (k=6) · Silhouette: 0.3029
Terroir Clusters
Layer 2 TF-IDF → PCA → K-means (k=7) · Fully model-derived · Silhouette: 0.2457

Reading the scatter plot

Each dot is a wine region. Proximity means similar identity profile — regions close together score alike on the six D-score dimensions. Colour marks cluster membership. Well-separated, cohesive colour groups indicate strong cluster structure. Overlap between colours shows where the boundaries are softer — particularly among The Moderates, whose defining feature is the absence of extremes. When the terroir panel is populated, compare the two: regions that are neighbours in identity space are often scattered in terroir space, and vice versa.