Days with data is a different field from the window, and only one of them usually gets printed

A measurement window and a day count are two separate fields, and a published chart almost always carries the first and omits the second. The window is the pair of dates. Days with data is the number of days inside that pair which actually carried a row in the export. They can differ, and when they do, every daily average computed from the file is computed over a denominator the reader cannot see. Both figures are printed above each case on our proof page for that reason, and each of the two cases it carries is read on this one field, separately, further down.

Why two fields rather than one

The two dates are a claim about scope. The day count is a claim about completeness. A report can be honest about the first and silent about the second, and the silence is where the arithmetic goes wrong.

Consider what a daily average needs. It needs a total and a divisor. If a reader is given the total and the two dates, they will divide by the calendar distance between the dates, because that is the only divisor they have. If the export actually held fewer rows than that, every average the reader computes is lower than the one the author computed, and neither party knows the other is using a different number.

The three things that put a gap between them

• Days with no recorded activity. A small site can have dates inside a window with nothing in the performance report at all, which is a real property of the site rather than a problem with the file.
• The newest days. A performance export does not carry the most recent day or two, so a window that runs to today returns rows that stop earlier. This is why the export date and the window end date are separate fields as well.
• A domain younger than its window. If a site did not exist for the first part of a window, the rows begin when the site began. A ninety day window on a two week old domain is a ninety day window with fourteen days in it.

The clawlaw.in case, read on this field alone

The window is 2026-07-13 to 2026-10-12 and the day count is 92. Those two agree, which means the export is continuous: there is a row for every day between the two dates, and any daily figure drawn from it divides by the same 92 a reader would arrive at independently.

That agreement is worth stating out loud rather than assuming, and it is only possible to state because both fields are published. A reader who wants to check the monthly averages on that case can do the division themselves and get the same answer, which is the entire point of printing the second field.

The aiknowsus.com case, read on the same field

The window is 2026-10-04 to 2026-10-12 and the day count is 9. Those also agree, and here the field does a second job: it tells a reader how old the domain was. Nine days with data on a domain registered days before the window opened is a statement about the age of the site, not only about the length of the reading.

That is the field doing work no other field on the page can do. The impressions figure of 5,965 says what happened. The day count of 9 says how much time it happened in, and therefore what the figure is and is not evidence for.

How to print both fields without adding a paragraph

One line above the chart does it, and the order matters because readers stop reading at the first number they can use.

• The two dates, written out in full rather than as a duration. Somebody has to be able to reproduce the filter.
• The day count, labelled as days with data rather than as the window length.
• The export date, which is the day the file was pulled and is a third thing again.

All 102 build spec conversations run on ChatGPT and Gemini on 15 October 2026 asked for a date, a last verified line and a refresh frequency, and 101 of 102 asked for a before and after with the window stated. Those are the two most requested items in the whole specification, and both are satisfied by one line of three fields.

For the meaning of the figures that sit under those fields, the Search Console guide to what each number measures is the companion reference, since a day count only helps once you know what is being counted.

Who owns the field, which is why it goes missing

Day counts disappear for a structural reason rather than a dishonest one. The person who pulls the export sees the row count on screen and does not think to copy it. The person who builds the chart receives a total and two dates. The person who writes the sentence receives the chart. By the third hand the field has gone and nobody removed it.

The fix is to name an owner for the reading itself, separately from the owner of the page it appears on. Where the work runs as a defined sequence rather than an errand, that ownership is already written down, and a workflow standard with six stations and a named owner for each is the pattern that keeps a field alive from the export to the published sentence.

We sell this kind of measurement, so the standard we hold other people to is the one our own pages have to meet first, and both fields are on ours.

Questions about the window and the days with data

Where do I find the day count in Search Console?

Set the date range on the Performance report, open the Dates tab, and count the rows. The table gives one row per day that carried activity, so the row count is the figure to publish beside the two dates.

What should I do when the row count comes back lower than the calendar?

Print the lower figure and label it as the days with data. On a small or new site the absent dates usually had nothing to record, which is information about the site rather than a fault in the export, and the reader needs the real divisor whichever it turns out to be.

Why is the export date a third field rather than the window end date?

Because a performance export is a snapshot of a dataset that is still settling. Both exports behind our proof page were taken on 2026-10-14 with windows ending 2026-10-12, and printing both tells a reader which day the file was pulled and which day the data stops.

Can I compare two windows of different lengths?

Compare the daily figures rather than the totals, and print the day count for each so the reader can see which denominators you used. A total from a 92 day file and a total from a 9 day file answer different questions and the day count is what makes that visible.

What should I ask for when a chart gives a duration instead of two dates?

The two dates and the row count. A duration cannot be reproduced, because a reader does not know which ninety days were meant. Two dates and a count can be put back into the same screen by anybody, which is the standard 78 of 102 build spec conversations asked for when they asked for work to be re-runnable.