The Date That Was Always Today

Our statistics page carried a line reading “Last updated” followed by a date. That date was always the day you happened to be reading it.

It was not a record of anything. The page asked the server what today was and printed the answer, on every single request, and it had done so since the page launched. A reader checking whether our numbers were current was told yes, always, by a line that could not have said anything else. The same value went into the machine-readable description of the page, which is the version search engines and AI systems read, so the claim was made twice: once to a person and once to a crawler.

The numbers on the page were live and correct the whole time. That is what makes this worth publishing rather than quietly fixing. A page can be completely accurate and still tell you something false about itself, and a freshness signal is exactly the kind of claim a reader cannot check.

The line now shows the most recent date on which the underlying data actually changed, taken from the counters that record it. When none of them can be read, the line does not appear at all. A page with nothing honest to say about its own freshness should say nothing.

Reading that page closely turned up three more things, all the same disease.

The count of published products appeared twice on it, from two different pieces of code. Both were correct, and nothing kept them that way. They now come from one function, the same one The Bar uses for the same figure.

The score ranges were built from a narrower set of products than the ranges on the page that defines those scores: this page skipped any score recorded as zero, and that page did not. A zero is not a low score, it is a data error, and the page whose job is to show our data was the one hiding it. Both pages now read one population, and a zero would show up on both.

The weights that combine the four dimensions into one score, and the total a product must reach to publish, were typed into the page as text. They were right. They were also unable to notice if the scoring code ever moved without them. They are now emitted by that code.

None of this changed a single number a reader saw. It changed what stands behind them, which is the part we can be held to.