Our robots keep a register of domain names and hosts. Each name enters through a source, it is probed, it is read, and its record is written. As of November 13, 2026, this register contains 2,734,407 known names, of which 2,386,817 have been probed, i.e. 87.3% of the total. A name never probed is not data, it is a line in a spreadsheet, so we do not count it as a site. Of the 2,386,817 names probed, 995,416 respond, i.e. 41.7%. The others do not respond, refuse automated visits, or no longer have an address. This distinction between known, probed and live is not a methodological nicety, it is what prevents announcing figures that mean nothing. It is a written rule, and we publish no account without its denominator. These three numbers read like the three floors of the same building. One can know a name without having visited it, and one can visit a name without getting a response. Each floor keeps its own total, and it is this total that gives meaning to the percentage. A figure without a denominator is an opinion dressed up as statistics, and that is exactly the kind of sentence this site refuses to publish.
The three figures that matter
- 2,734,407 known names, everything our sources have provided, grafts and directories included.
- 2,386,817 domains probed, i.e. 87.3% of the register. A name never probed is not data.
- 995,416 live sites, i.e. 41.7% of the domains probed. The others do not respond, refuse, or no longer have DNS.
A live site is not a known site. Between the two, there is a reading.
What we really know about live sites
Of the 995,416 sites that respond, only two measurements are advanced enough to be published without reservation. 149,265 sites have a read core version, i.e. 15.0% of the live ones. 14,489 sites have a complete compliance check, i.e. 1.5%. Everything else is a work in progress, and this work has a name, it is reading. An unread version is a vulnerability that cannot be attributed, hence an alert that cannot be sent to its owner. An unchecked compliance is a legal obligation whose state nobody knows, and it is not at the time of an inspection for the purchase of a cooking robot that it can still be corrected. We prefer to publish these two low percentages rather than dress them up with estimates, because an estimate cannot be verified and neither can a client. The most embarrassing figure in this article does not look like progress, it looks like a silence. Among the live sites, 1,518 publish no version, neither in the header, nor in the page, nor in their public files. They are neither healthy nor sick, they are unjudgeable, and we write that word as is instead of filing them on the side that suits us. The details of this silence are told in the article on the Apache vulnerability of November 11, where these sites were counted one by one rather than assumed.
The three steps of a record
- The name enters through a source, with its date.
- We read what the site publishes, and nothing more.
- The record is calculated, and each figure carries its denominator.
Why do we not count sites we cannot read ?
Announcing a percentage of vulnerable sites across the entire estate would be easy and it would be wrong. A site that has not published its version is not a healthy site, and it is not a sick site either, it is a site about which nothing can be said. Classifying it on the side that suits us would make a better headline and a worse measurement. Out of the 995,416 live sites, the 1,518 that stay silent are therefore neither in the numerator nor in the denominator of our vulnerability rates, and that is why these rates always carry the mention of their calculation base. The same rule applies to legal compliance, where a missing legal notice page is not equivalent to an absence of notices. This requirement has a cost, it forbids us round numbers and punchy sentences, and it makes our pages less spectacular than those that announce millions of sites in danger. In exchange, every number published here can be re-read in the database and reproduced by anyone, which is the only meaning we give to the word measurement. The full catalogue can be consulted in the vulnerability records, with the share of each software and the detail of the version ranges.
