crawlora-net/dead-web-index

Dead-Web Index

A reachability census of the DomCop top 10 million domains (June 2026). Every domain is probed and labelled alive / redirect / blocked / dead twice — once by a polite plain-HTTP bot, and once by a reachability client with a real Chrome TLS/JA3 fingerprint. This is the dataset behind the Crawlora Dead-Web Index explorer.

Headline

• 14.1% truly dead — no DNS, refused/reset connection, nothing listening.
• 8.9% blocked but alive — a live server returning 403 / 429 / anti-bot.
• 76.6% alive, 0.3% redirect.
• A real Chrome TLS fingerprint reaches ~72,000 domains the polite bot cannot — the gap between "dead to a bot" and "alive to a browser".

What "dead" means

A domain is dead only if genuinely unreachable — no DNS resolution, a refused/reset connection, or nothing accepting a TCP connection. A server that answers any HTTP status (even 404 or 5xx) is alive or blocked, not dead.

Files

• data/polite.part-01..10.jsonl.zst — polite arm, 10 rank-partitioned chunks (zstd-compressed JSONL), 9,992,781 rows total.
• data/reachability.part-01..10.jsonl.zst — reachability (browser-fingerprint) arm, 9,997,315 rows total.
• data/sample.jsonl — 1,000-row uncompressed preview (both arms).
• data/manifest.json — per-chunk rows, rank range, byte size, and sha256.
• data/summary.json — run-level aggregates (outcome split, by-reason, by-TLD).

Decode with zstd -dc data/polite.part-*.jsonl.zst; DuckDB and Polars read .jsonl.zst directly.

Schema (one JSON record per line)

domain, tld, rank, mode (polite|reachability), outcome (alive|redirect|blocked|dead), reason, first_status, final_status, final_url, scheme, hops, parked, run_id, probed_at.

License & sources

CC BY 4.0. Source repo: https://github.com/Crawlora-org/dead-web-index-data · interactive explorer: https://crawlora.net/dead-web-index · methodology: https://crawlora.net/blog/how-much-of-the-web-is-dead-2026 · produced with the open-source crawlora-deadweb classifier: https://github.com/Crawlora-org/crawlora-deadweb