How do you detect when K-Means is completely failing on a dataset?

Failure is detected through low silhouette scores, unstable clusters, and high inertia without clear elbow.

If clusters overlap heavily, silhouette scores approach zero or negative values. If inertia decreases smoothly without elbow, structure may not exist. Repeated runs producing inconsistent labels also indicate failure.