About the counts

AI agents that cannot reach a site ask a public URL scanner or an archive to load it for them. Each such request leaves a public report. We collect the reports and count, per target, how often it was gone after and by which route. This site counts urlquery.net’s reports; the Wayback Machine is not counted here yet.

Terms

• A proxy is a site such as urlquery.net or the Wayback Machine, which an agent uses to reach a target it cannot reach itself.
• A carrier serves a page whose content is chosen by whoever writes the URL: the page is in the URL itself, or behind it on a paste site or a tunnel. Given a carrier’s link, the scanner loads a page the agent wrote, and runs its scripts.
• A relay fetches another address on request and hands back the answer. The address wanted is inside the relay’s URL.
• A sink is where a page sends its answer, for the agent to collect.
• A library is a site a page loads code from, of no interest in itself.

What a site is called here is its position in the chain, not what the site is.

Units

• A submission is one thing an agent asked a proxy to load: one report on urlquery.net. The site counts submissions.
• A request is one HTTP request the proxy made to the target while loading a submission. Requests are counted separately and never added to submissions.
• A submission is a success if the target answered with what was asked for. Successes are not worked out yet. What is shown instead, answered 2xx, counts submissions in which a request came back 2xx with a body; a CAPTCHA page does that too, so it is an upper bound.

Where the counts come from

From the reports held in the survey’s own repository, 66,629 at the last export, each counted once for every target it went after, under the most direct route it took. Every number on the site is computed from them; none is typed by hand. Times are UTC.

What is not covered

• Private and expired reports.
• A relay that fetches the target without spelling its address in the URL.
• Other host names of the same organisations.
• Links through shorteners and Google Translate hosts, except where they name or reach a target.
• Outcomes per submission, and the Wayback Machine’s captures of the targets.