What this site measures

This site has no trackers, no cookies and no consent banner, and it loads no fonts or scripts from any other host. It measures one thing, and this page says what.

Everything described here can be checked from your browser. The last section says how.

The only thing measured: a search that returned nothing

When a search on this site returns no results, the page can send the term you looked for. A search that finds something sends nothing: the site does not measure which searches are popular, it measures the gaps, to know which pages to write.

At the bottom of each tutorial, two buttons, “it worked” and “I’m stuck”, send something only if you click.

The four fields

Here is everything a search with no results sends:

{
  "terme": "passerelle hors ligne",
  "langue": "fr",
  "version": "1.0.0",
  "jour": "2026-08-05"
}

Field	What it is
terme	the term you looked for, lowercased and capped at 60 characters
langue	fr or en, the language of the page
version	the version of the published documentation
jour	the date, without the time

The time is left out: paired with a rare term, it would let two records be linked together. The day alone does not.

A click on a tutorial’s end-of-page button sends three fields: the name of the tutorial, the verdict (ok or bloque) and, for a blocker, the step picked from a closed list of four values. There is no free-text field anywhere on this site.

What is not collected

• The IP address. It arrives with every HTTP request and it is discarded before anything is written: not logged, not hashed, not truncated, not geolocated. A hashed address is still personal data; so is a truncated one, in a company where the block identifies the organisation.
• HTTP headers: no User-Agent, no Referer, no Accept-Language, no Cookie. The only header read is Content-Length, to reject an oversized body before reading it.
• An identifier: no cookie, no read or write in local or session storage, no browser fingerprint, no random number, not even a short-lived one.
• A journey: not the page you came from, not the sequence of your searches, not their order. Two searches made by you are indistinguishable from two searches made by two people; that is what makes the measurement non-personal.
• Any link with anything else, neither with the installed product nor between two measurements. That would require a shared identifier.

What your browser refuses to send, on your workstation

The filter that decides whether a term should be sent runs on your machine, before any network call. A rejected term does not leave your workstation. You can check this filter by reading the code the page sent you, unlike filtering done on a server.

The following are rejected, with no notice to you:

What is rejected	What it protects
more than 60 characters, or more than 8 words	a pasted error message
a @	an email address
a ://	an internal address, a token
a file path, including Windows paths	the directory tree of your workstation
an IP address	your network topology
a run of 16 characters or more that looks like a key or token	a key, a token, a fingerprint
any character outside letters, digits, space and - _ . '	everything else

Someone who is stuck often pastes an error message into a search box. In a defence or banking environment, that message carries an internal host name, a customer reference, sometimes a token. The site does not receive it.

Why there is no cookie banner

• Nothing is written to or read from your device: no cookie, no local storage, no fingerprint. The legal trigger for storage consent does not apply.
• No personal data is kept: the IP address is discarded on arrival and no identifier exists, so nothing ties a record to a person.
• If your browser signals that it does not want to be tracked (DNT or Sec-GPC), the site sends nothing.

This page describes a technical stance and the controls that hold it. It is not legal advice.

What becomes of these terms

Terms are counted and ranked by frequency, over a rolling 90-day window, after which the raw records are deleted. A term seen only once does not appear in the report.

The result is a list read every week to decide which pages to write. It is not published: a publicly readable collection point is a surface for pollution.

How to check it yourself

1. Open your browser’s network tab, then search for a term that exists on this site. No request goes out: a search that finds something sends nothing.
2. Search for a term that cannot be found. A single request goes out, and you can read its contents: the four fields above. Search for the same term a second time without reloading the page: nothing goes out again.
3. Paste an error message containing an internal address or a token. No request goes out.
4. Look at your cookies and your local storage in your browser’s dedicated tab. Nothing from this site is there, before or after a search.

This site’s search runs entirely in your browser: the index ships with the pages, and no server sees what you type. The measurement described here is therefore explicit: no server log could produce it.

The state of collection

Collection is live. The address that receives these terms is written in plain text in this page’s source code, under the data-mesures-recherche-url attribute.

What receives them is a service run by the publisher, at the same European host as the gateway. No third-party analytics service is involved.

That service can only write. The account it uses has no right to read back or delete what it has stored, nor to connect to another database. What it stores is pulled back every week to a workstation at the publisher and deleted from the host in the same transaction: nothing stays there longer than a week.

The documentation delivered offline sends nothing. The package an air-gapped organisation receives is built with no collection address: the sending code is not in it. You can confirm this absence by searching for data-mesures-recherche-url in the package.

What is not measured

The site publishes no tutorial completion rate. A rate means knowing how many people started, so recognising that the visit at the start and the visit at the end come from the same person, so having an identifier. The two end-of-tutorial buttons give a count of voluntary replies, not a rate.