emperor-mew/voidly-anomaly-domain-drift-v1

Voidly Atlas Domain Drift HDBSCAN v1

Version: v1 | Trained: 2026-05-21T04:43:38.216279Z | License: CC BY 4.0

Weekly per-domain HDBSCAN drift surface. Variable-density clustering over the last-28-day feature vector for every domain with >=10 measurements. Orthogonal axis to per-country DBSCAN.

Eval

Metric	Value
algorithm	HDBSCAN(min_cluster_size=5, metric=euclidean, after StandardScaler)
min_cluster_size	5
n_domains_this	27
n_clusters_this	2
n_new_clusters	2
n_high_drift_clusters	0

Features

• log_n
• avg_block_rate
• std_block_rate
• n_countries_blocking
• asn_unique_log
• source_diversity
• pct_dns_block
• pct_tcp_reset
• pct_blockpage
• pct_tls_reset
• pct_isp_outage
• cat_NEWS
• cat_ANON
• cat_GRP
• cat_PORN
• cat_MSG
• cat_SRCH
• cat_OTHER

Honest caveats

• Only 27 domains pass the min-10-measurement filter — week-over-week cluster stability needs months more data.
• Drift score is L2 distance in this-week's standardized space; domains absent from one week get status=new_this_week / dropped_this_week instead of a score.
• ORTHOGONAL to per-country DBSCAN (voidly-anomaly-dbscan-v1) — different axis, different surface.

Citation

@misc{voidly_voidly_anomaly_domain_drift_v1,
  title  = {Voidly Atlas: voidly-anomaly-domain-drift-v1 (v1)},
  author = {Voidly},
  year   = {2026},
  url    = {https://huggingface.co/emperor-mew/voidly-anomaly-domain-drift-v1},
  note   = {Open censorship-research ML stack. CC BY 4.0.}
}

Method foundation: McInnes et al. 2017 — arXiv:1705.07321