crawlora-net/dead-web-commoncrawl

Dead-Web Common Crawl — a longitudinal host-reachability panel (2018–2026)

 ·  Hugging Face  ·  Kaggle  ·  License: CC BY 4.0

An open dataset labeling the reachability of 172,959,928 registered domains across 80 monthly Common Crawl archives (CC-MAIN-2018-05 → CC-MAIN-2026-21), built from Common Crawl's robotstxt subset and calibrated against a live re-probe to separate genuinely-dead domains from those merely blocking crawlers or dropped from the crawl.

Headline

• ≈ 40 million domains (~23% of the 8-year union) are calibrated genuinely dead (no DNS / nothing listening).
• Disappearance from Common Crawl over-counts death: of domains that vanished, only 64.2 % [63.5–64.9 %] are truly dead — 31 % still resolve (dropped/blocked but alive), 4.7 % are blocked.
• The dead-vs-blocked distinction (which prior link-rot work conflates) is measured here: domains last seen blocked are only 44.7 % dead — 46 % are alive, just refusing crawlers.

Method

1. Extract — each crawl's robotstxt subset → one row per registered domain (presence + HTTP-status mix). fetch_status is HTTP-only, so a truly dead host produces no row; death is inferred from disappearance across crawls.
2. Label — per domain: first_ord, last_ord, total_crawls (80), n_present, last_state, recent_absence_streak, and a label: alive_present · dead_candidate (established then went dark from a healthy state) · dark_ambiguous (went dark from blocked/erroring) · intermittent (sparse coverage — not counted as death).
3. Calibrate — a stratified 20 k sample of disappearance candidates was live-re-probed (DNS + TCP + HTTP); per-stratum dead rates (Wilson 95 % CIs) convert raw "disappeared" counts into a defensible "% truly dead." Absence ≠ death — calibration is the point.

Label distribution

label	domains	%
intermittent	65,778,102	38.0
dead_candidate	55,525,600	32.1
alive_present	44,246,249	25.6
dark_ambiguous	7,409,977	4.3

Calibration — of domains that disappeared, what's truly dead

last-seen state	truly dead (95 % CI)	blocked	alive (false-positive)
overall	64.2 % [63.5–64.9]	4.7 %	31.1 %
ok (healthy)	70.7 % [69.8–71.6]	3.4 %	25.9 %
servererror	76.8 % [73.2–80.0]	3.1 %	20.1 %
notfound	60.7 % [59.3–62.1]	5.3 %	34.0 %
redirect	55.7 % [53.8–57.5]	5.8 %	38.5 %
blocked	44.7 % [42.4–47.1]	9.2 %	46.0 %

Files

• panel.part-00..31.parquet — full per-domain panel (172.96 M rows, zstd parquet) — release assets.
• data/sample.jsonl — 1,000-row preview · manifest.json — per-part rows + sha256.

Built with

The open-source crawlora-deadweb CLI (cc subcommands). Companion to the Crawlora Dead-Web Index.

Mirrors

• Zenodo (archival, DOI): https://doi.org/10.5281/zenodo.20814397
• Hugging Face: https://huggingface.co/datasets/crawlora-net/dead-web-commoncrawl
• Kaggle: https://www.kaggle.com/datasets/plantpark/dead-web-commoncrawl

Cite this dataset

Please cite the concept DOI 10.5281/zenodo.20814397 (always resolves to the latest version; this release is v1.0.0, DOI 10.5281/zenodo.20814398):

Crawlora (2026). Dead-Web Common Crawl: a longitudinal host-reachability panel (2018–2026) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.20814397

@dataset{crawlora_dead_web_commoncrawl_2026,
  author    = {Crawlora},
  title     = {{Dead-Web Common Crawl: a longitudinal host-reachability panel (2018--2026)}},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.20814397},
  url       = {https://doi.org/10.5281/zenodo.20814397}
}

License

Data: CC BY 4.0. Source: Common Crawl. See LICENSE · CITATION.cff.