Data

The whole census, every column, one row per server — plus the per-link evidence behind every jurisdiction claim.

Census CSV · Parquet · Jurisdiction links CSV · summary JSON

Run 20260910T170344-classify-42c3a5 — 1,273 servers, 28 jurisdictions.

Recomputing the weighting

jurisdiction-links.csv is one row per (server, country, basis, weight, reason). If you disagree with the weights — and you might reasonably think a registry record should outrank a declaration, or that language should count for nothing — you can regroup that file and get a different answer without re-crawling anybody.

Tables beyond the census

table	what it holds
`jurisdiction_link`	one row per server, country and basis — the evidence
`regime_signal`	which regulators a server has noticed (OSA, DSA, NetzDG, DMCA...)
`activity`	twelve weekly buckets of posts, logins and registrations
`registration`	RDAP: registrar, dates, nameservers, redaction status
`cert_name`	hostnames observed in Certificate Transparency
`rule`, `policy_text`	published rules and policies, hashed so changes are visible
`probe`	every HTTP request made, with status and timing — the provenance trail

Licence and attribution

The census is original measurement and free to reuse with credit. It derives in part from DB-IP's IP-to-country data, which is CC BY 4.0 — if you redistribute these files, carry that notice.

Corrections

If your server is described wrongly, the fix is usually in what it publishes: check your /api/v2/instance and your rules. If the hosting resolution is wrong — a stale IP, a CDN we failed to detect — that is our bug and worth reporting.