Method.
A measurement is only a measurement if someone else can produce the same number. This page states what was measured, how, and what was not.
Sample
The list was not chosen but filtered. The source is an independent ranking; the filter is the country domain. This means the list carries none of our preferences.
Probes
At most three requests of our own per site, plus one request per probe. No page is crawled and no content is stored; only headers and countable facts are recorded.
The identity is not hidden
The agent name is preserved in the probe string, because server rules match on that name; the address of this page is appended, because whoever measures should be visible in the log. Imitation alone would make the measurement valid but hide the identity; our own name alone would reveal the identity but invalidate the measurement.
The placebo probe
The control probe is a browser; the measured agents are not. On a server that turns away every non-browser request, measuring with the control alone would mistake that behaviour for something agent-specific. So one more probe is sent, shaped like an agent but not on the list. If it also receives a different response, the behaviour is general and the row is not listed.
Limits of the measurement
A single run measures a moment, not a distribution; that is why the time series is kept. The measurement is made from one geographic location, and delivery networks may decide differently by region. A status code alone does not show intent: configuration, a rule template or a third-party layer can produce the same code.
Reproduction
The measuring tool is open source and the same command filters the same list. If a different number appears, the difference is either in the date or the location; both are recorded.
Objection
If you believe a row is wrong, write to us: the measurement is repeated and whatever comes out is published. A correction is recorded too, never made silently.