SPORE clustering algorithm lands paper, package

This is a thoughtful attempt to solve the classic density-clustering failure mode: either merge nearby clusters or fragment irregular ones. The two-stage design is the interesting part, because it lets SPORE grow conservatively first, then sharpen boundaries later instead of forcing one pass to do both jobs.

• –The expansion phase is essentially an adaptive, statistics-gated region grower, which should help on nonconvex clusters where DBSCAN-style methods struggle.
• –The SCR reassignment pass is the practical differentiator: it can clean up boundary fragments after the algorithm has already isolated cluster cores.
• –Approximate k-NN via HNSW makes the approach more usable at scale, especially if the clustering logic only needs rough neighborhood structure.
• –The main question is generalization: the benchmark claims are promising, but the method will need independent tests against HDBSCAN, spectral clustering, and modern graph-based baselines.
• –The PyPI release makes this more than a paper artifact, which matters if the implementation is stable and the defaults are sane.