# src/kmeans
## Purpose
K-means++ clustering: smart initialization + iterative Lloyd refinement.
Used by spectral clustering backend.
## Surfaces
- `kmeans_pp(embeddings, k, max_iter) -> Vec<usize>`
## Dependencies
None (self-contained).
## Invariants
- Output labels are in 0..k.
## Verification
```bash
cargo test --lib kmeans
```
## Notes
- Uses Euclidean distance on f32 vectors.