Update cadence & snapshots
webrank is refreshed on two clocks so you get both freshness and citability.
Daily
Every day (~03:20 UTC) we re-fetch the daily-cadence sources — Tranco, Majestic, Cisco Umbrella, Cloudflare Radar — and recompute both numbers (webrank + pagerank) across the whole domain set, then rebuild the fast lookup and re-export the downloadable list. So the live API and webrank-latest.csv.gz are never more than ~24 h old — fresher than a monthly list.
The monthly-cadence sources (Chrome UX Report, Common Crawl, OpenPageRank, host.io) refresh on their own schedule; the daily compute always uses the latest available snapshot of each, so they flow in automatically the day after they publish.
Monthly — permanent, citable snapshots
On the 15th of each month we re-fetch the monthly sources, re-derive the source weights on the fresh data, do a full recompute, and freeze the result as a permanent, immutable list:
webrank-2026-07.csv.gz # the July 2026 snapshot — never changesThese are the lists to cite (papers, reproducible research, audits): unlike latest, a dated snapshot is byte-stable forever. Available at api.webrank.top/list/webrank-YYYY-MM.csv.gz and mirrored to GitHub (webrank-top/webrank-lists) so every month has a permanent public home. Each monthly release also ships the full list (every ranked domain, ~146M rows) as split gzipped parts — self-hosted at webrank.top/downloads and mirrored to GitHub Releases, and archived by Zenodo with a permanent DOI — see cite. Everything stays reachable on our own site even if GitHub is down.
What "latest" means
| file | contents | changes |
|---|---|---|
webrank-latest.csv.gz | the current live top-1M | daily |
webrank-YYYY-MM.csv.gz | that month's frozen snapshot | never |
Every API response and list carries the algorithm version (currently v4), so results stay reproducible across method changes — a new method ships as a new version, old snapshots keep the version that produced them.
Weights
Source weights are re-derived monthly (source reliability moves slowly), on the fresh month's data, and versioned. The current weights and the validation numbers are on the methodology page.
Changelog
- v4 (2026-07-11) — pagerank folds in referring-subnet diversity (our own
measurement: distinct /24 networks among a domain's linkers, derived authority weight 0.96): network-concentrated link farms drop 10–15×, webrank (traffic) unchanged (v3→v4 correlation 0.9998). Monthly trends stitch v3 history seamlessly. Details on the methodology page.
- 2026-07-11 — monthly snapshots get automatic DOIs (Zenodo); full ranked list
(~146M rows) published per month as GitHub Release assets; official clients: pip install webrank, npm i @webrank/webrank; /history API sped up ~20×.
- 2026-07-10 — public site launch: methodology + docs, live lookup, leaderboards,
per-domain pages, top-by-TLD lists, sitemaps. Daily refresh automated.
- v3 (2026-07) — two data-driven numbers:
webrank(traffic-anchored popularity) +
pagerank (link-anchored authority). Weights derived from agreement with real-traffic sources; predicts real user traffic ~46% better than equal weighting. 146M domains ranked of ~293M tracked (see coverage).
- v2 (earlier) — single composite over 13 months of history; cut month-to-month churn ~45%
via 3-release Common Crawl smoothing.