A creator sees a completion figure in a dashboard.
It looks like a verdict.
It has the clean shape of one.
But a completion rate says nothing about whether a reader finished the piece until the report says what counted as finishing.
One article can fire completion when the reader passes halfway down the document.
Another can fire when the final paragraph comes into view.
Both call the result completion.
They are not measuring the same exit.
Scroll depth measures page height.
An end marker measures a place in the page.
A long comment section or a heavy sidebar changes the first one without touching the article at all.
Which is why two reports showing the same number can’t be compared.
The percentage hasn’t settled the question.
It has named the question still worth asking.
So start with the event.
Name the population that began. Name the threshold that ends the count. Name whether the trigger follows document height, a specific element, or something else entirely.
The browser infers none of this.
A script fires when the thing you configured happens.
A scroll past the halfway mark is evidence that the halfway mark was crossed.
It is not evidence that half the meaning arrived.
Video makes the distinction sharper.
Its familiar quarter-marks record playback position, nothing more.
An autoplaying player in a background tab can reach the end of a video honestly, and without a visibility check the report will count it.
That doesn’t make completion useless.
It makes the definition part of the metric.
The same restraint belongs to dwell time.
Time on page is usually measured from one hit to the next, and the last page of a visit has no next hit.
So the final page can be recorded as zero, or dropped entirely, unless it sends its own signal on the way out.
Which means the page people read carefully before leaving can look like the page that held them least.
Foreground-time measurement improves the picture.
It still depends on periodic check-ins, so a reader who stops scrolling for a long stretch can vanish from the clock while still reading.
Dwell is useful for finding a page worth inspecting.
It is not a measure of satisfaction.
None of these numbers are, and that is fine — as long as you know which one you are holding.
Then separate the work that qualifies a page from the work that learns from it.
The check at publication asks whether the page shipped correctly. Byline, image, links, metadata, canonical tag.
That is pass or fail against the thing that went live.
The analysis comes later, and asks something else. Where did people arrive from, which event fired, where did they stop reaching the next one, and what decision put the break there?
Don’t collapse either into a quality score out of five.
Use pass-or-fail checks for what can be checked. The event exists. The end marker was placed deliberately. The canonical tag resolves. The claim has a source.
Keep a note for everything that doesn’t belong in a checkbox.
Then let the behavior lead back to a decision you can actually make.
If completion falls after a new opening, look at the promise that opening made.
If it falls after a section moved below a block of related posts, look at the route through the page.
And if the number only changed after the page grew a long footer, look at the ruler before you rewrite a word.
That last one is the common case, and it is the one that costs people a rewrite they never needed.
So define what the reader began.
Define what it means to reach the end.
Then put the event beside the part of the piece you can change.
The completion figure can start that work.
It can’t finish it.
A metric becomes useful only when it points back to a decision you can change.