Ask k-means for a number of clusters and score every point by how much closer it is to its own cluster than to the next one. Change the true number of groups, how far apart they sit and the k you request, and see the per-point silhouette, the mean silhouette against k and why the elbow of the inertia curve is not enough. A Blender-built plotting easel with a hand-computed example, scale invariance, range and Lloyd monotonicity checks built in.