Pageviews

A pageview is sent when the tracker starts and again on every client-side navigation. It is the only event most sites ever need.

What goes out

The payload is deliberately tiny (one and two character keys) because it rides inside a sendBeacon and the tracker’s whole job is to be small.

{
  "s": 1,
  "t": "pageview",
  "u": "https://example.com/docs/x?ref=hn",
  "r": "https://news.ycombinator.com/",
  "w": 1512, "h": 982,
  "vw": 1280, "vh": 720,
  "l": "en-GB",
  "tz": "Europe/Lisbon",
  "d": 4210,
  "sc": 68,
  "v": "0.1.0"
}

Everything empty is left out rather than sent, because every column it maps to already carries that default.

The user agent is not in there. The server reads the request header it was going to receive anyway, and derives browser, OS and device type from it. Nor is the IP: the server resolves it to country, region, city and ASN, folds it into the visitor hash and then drops it. No raw address is ever stored on a human event row.

How the URL is read

The full URL goes up; the server splits it. The hash is stripped unless you set data-hash, because #section is usually a position on a page rather than a different page. The query string is kept whole, and its parameters are also stored as a map, so utm_source and friends can be filtered on by name.

Campaign parameters (utm_source, utm_medium, utm_campaign, utm_term, utm_content) and click ids (gclid, fbclid, msclkid, ttclid) are lifted into their own columns.

Sending one by hand

micaforge.pageview();
micaforge.pageview({ url: "https://example.com/checkout/step-2" });
micaforge.pageview({ referrer: "https://partner.example/", props: { variant: "b" } });

A URL you pass does not have to exist. Reporting a step in a dialog as a virtual page is a legitimate use, and it keeps funnels readable.

To stop the automatic one and send every pageview yourself, set data-auto-pageview="false".

Engagement time and scroll depth

Neither is sent with the pageview it belongs to, because neither is known yet.

The tracker runs a clock while the page is visible and stops it when the tab is hidden. It records the furthest point down the page the reader reached. When the page is hidden or unloaded, both go out as one engagement event; when the reader navigates inside a single-page app, they ride along on the next pageview as d and sc.

So duration_ms on a pageview row is the engagement time on the page before it. That is the only way to measure it without a second request per page, and it is why the last page of a session has a duration and the first does not.

Time in a background tab does not count. A reader who opens twelve tabs and reads one has one page with a duration.

Sessions

A session is a visitor’s activity with less than 30 minutes between hits. The window is a constant, not a setting: a site that quietly redefined a session would report numbers that could not be compared with anyone else’s, including its own from last month.

What is never reported

Paths matched by data-exclude are dropped in the browser, before anything is sent. localhost, 127.0.0.1, ::1 and any .local host are suppressed unless data-debug is on, because a developer’s own refreshes are not data. And a browser that has called optOut() sends nothing at all until optIn().