Why a proof page carries its export date and not just its window

A window and an export date answer two different questions, and a page that prints one of them has answered half. The window tells a reader which days are inside the number. The export date tells them when the number stopped being revised, which matters because Google revises recent days for a while after they occur. Both exports behind our proof page were taken on 2026-10-14, and that date is printed beside each window rather than left in a file name. Without it a reader cannot tell whether they are looking at settled figures or at the last few days of an export still in motion.

What revision actually does to a figure

Search Console reports recent days before the processing behind them is complete, so the most recent rows in any export can move after you take it. The direction is usually upward as late data arrives, and the size of the movement is largest on the final day or two. That has one practical consequence: the last day of a window is the least trustworthy row in it, and a window that ends on the day of the export is making its weakest row the most prominent one.

Our own windows end two days before the export was taken, on 2026-10-12 against an export of 2026-10-14. That gap is not decoration. It is the reason the final figures in each series can be quoted at all.

The four dates a published figure can carry

Date	What it answers	What goes wrong without it
Window start	Which days are counted	A total with no scope
Window end	Where the series stops	An end date that might have been chosen
Export date	When the data stopped moving	Unsettled rows quoted as final
Publication date	How old the claim is today	A stale figure reading as current

Four dates, and most published charts carry one or two. The expensive omission is the third, because it is the only one a reader cannot infer or guess at from the chart itself.

The engines ask for this directly

We put this question to ChatGPT and Gemini on 15 October 2026, across 102 conversations: shown an answer the engine had just produced, what would a page need to publish before it got used as the source instead. The most requested item came back unanimously, and it is three fields rather than one. A publication date. A line recording when the facts were last checked. A stated interval for checking them again. A page carrying the first of those has met a third of the thing all 102 conversations asked for.

Those are three different promises, and a reader can tell at a glance which of them you have made. The first is about authorship, the second about maintenance and the third about intent, and only the second and third say anything about whether the page is still true today.

Our own restated facts have gone stale inside two days

This is not a hypothetical risk and the clearest example is ours. Reading OpenAI's crawler documentation on 16 October 2026 turned up a fourth identity, and two pages of ours were still describing three. One of them put the count in a sentence, so it had gone from accurate to false without anybody touching it. Both now carry the fourth name with the date the list was read. Neither page was careless when it was written. It restated a document somebody else controls, and the document moved.

So the rule we hold to is that any page restating an external list carries the date it was read, and the station in the workflow that catches this kind of drift is a named step rather than a good intention. The workflow standard we publish puts the date check and the contradiction sweep at stations five and six, after every wave, including work done an hour earlier.

How to date a figure so it survives being checked

• Print the window as two dates rather than a length. Ninety two days is not a window; 2026-07-13 to 2026-10-12 is.
• Print the export date beside it, and end the window a day or two earlier than the export.
• Add a last verified line that is a different field from the publication date, and update it when somebody actually looks.
• Say how often the figure gets re-read. A stated interval is a promise a reader can hold you to.
• When a restated fact changes, say on the page that it changed and when, rather than editing it quietly.

Questions about export dates and Search Console revisions

How long does Search Console keep revising a day?

The most recent days move the most, and the movement settles after a short lag. The safe habit is to treat the final day or two of any export as provisional and to end a published window before them.

If I re-export the same window later, will the numbers match?

The older days will. The days nearest the original export may have risen. That is exactly why the export date belongs on the page: it tells a reader which version of the window they are holding.

Is a publication date not enough?

It tells a reader when the page was written and nothing about the data inside it. A page published in October can carry a window from July and an export from August, and all three of those facts are needed to interpret the figure.

What should a last verified line say when nothing changed?

The date somebody checked, and that the figures were unchanged. A check that found nothing is still a check, and recording it is what distinguishes a maintained page from an abandoned one.

Why would a vendor make this much of a date field?

Because measurement is what we sell, along with the page writing behind it, and a measurement with no date on it is an opinion with a number inside. Asking buyers to hold suppliers to a standard means publishing our own figures against it first.

Two dates, and the one that makes the other usable

Of the two, the window is the one everybody prints and the export date is the one that makes the window mean anything. A window tells you what was counted. The export date tells you when the counting stopped, and without it the most recent and most interesting part of any series is unreadable. It costs one short line of text. The reason it is missing from almost every chart in this trade is not expense, it is that nobody asks for it, which is a very good reason to start asking.