What this site measures

Written by a person. Last read by a person on 2026-09-06, 22 days ago. Its facts were checked by the eval suite on 2026-09-28.

You want to know what this site records about you, you are deciding whether to trust a claim it makes about reader outcomes, or you are building something similar and want to see where the honest limits of documentation measurement sit.

What is recorded is listed below. Whether anything is recording is answered by the counters themselves, not by this sentence. A page that states its own state in prose goes stale on the day that state changes. That is what happened to the sentence this replaces. It said nothing was being measured, and it would have gone on saying so after a database was bound.

This page was written before any data arrived, on purpose. A measurement page written afterwards tends to describe whatever was convenient to collect.

What will be recorded

Aggregate counts, and nothing else. Each is a number that goes up.

Which route a stuck reader took. Whether they arrived at a troubleshooting section from the task list, the help page, or a search engine, and whether they left it for the canonical page or for another troubleshooting section.

A troubleshooting section visited more than once in a session. Reading the same section twice is a signal that it did not answer the question the first time.

A faded example reopened after being closed. The reader tried it alone and came back.

Where the classifier sends people who then bounce. A destination readers leave immediately is a routing error, and it is invisible without counting.

The numbers

There is no view of one reader here because there is no such view to build. The table holds a signal, a dimension, a date and a count, and nothing else, so a total is not a summary of something finer that is being withheld: it is the whole of what was kept.

See it for yourself

Every page carries a control at the bottom reading What would this page send?

It opens a panel showing three things: the exact body that would be posted if you asked the reading helper a question, everything this site is keeping in your browser right now, and what is being counted about your visit. Nothing is summarized. The payload is shown as the payload, after personal data has been removed, and you can delete any of it before it goes anywhere.

The panel reads the same code that does the sending rather than describing it separately. That matters more than it sounds. A privacy page maintained apart from the thing it describes tells a comfortable lie the first time the sender gains a field and the page does not, and the page is the one a reader believes.

This is not a setting and it is not a promise. It is the behavior, displayed, and it is the only claim on this page you do not have to take from us.

What is not recorded

No identifiers. No session tokens. No cookies. No fingerprint. No timestamps precise enough to correlate two visits into one person's path.

No third-party anything: no analytics service, no tag manager, no font host, no content network. The footer of every page says so, and there is nothing to verify because there is nothing to load.

If a signal cannot be recorded as an increment to a shared counter, it is not recorded. That rules out several things worth knowing, and this page names them rather than quietly dropping them.

What cannot be measured here

The four-level framework most organizations use for training outcomes reaches further than this site can.

Level 1, reaction. Reachable. Whether readers can use a page shows up in the route counters above.

Level 2, learning. Partly reachable. Whether a reader can now do the thing is what the concept criteria and the example checkpoints are for, but nothing here observes a reader meeting them. A self-check they run themselves is the closest honest instrument, and it is Deliverable C.

Level 3, behavior. Not reachable from here. Whether a reader writes a bounded retry loop next week happens in their codebase, not on this site. It would have to be self-reported, and inference from page counts is not evidence. This page will not claim it.

Level 4, results. Not reachable in a way that respects the reader. Tying documentation to support load or incident rate needs data from systems this site cannot see, joined per person. That join is the thing being refused. The level is named here rather than omitted, because a framework quietly missing its last two levels reads as a framework that was met.

Signals that are not evidence of use

Exports and stars get counted, and they are reported as what they are: somebody found this worth keeping, or worth marking. Neither says a page taught anybody anything.

The temptation is to report them as engagement, because they are the numbers that go up and they are the cheapest ones to collect. A documentation project that measures what is convenient and calls it outcomes is doing the thing this whole system exists to argue against.

Why this page exists before the data

A measurement plan written after the numbers arrive is written around them. Writing it first means the limits are stated by somebody who does not yet know which of them will be inconvenient.