Our Lab Shelves Now Say the Same Thing to Everyone

Every shelf in our Lab carries a sentence saying what it holds. Until today each of those shelves carried three of them, and no two matched.

One sentence was written for the reader and appears at the top of the page. A second was written for search engines and appears in the result you see on Google. A third was written for machines that read the page’s structured data. Across the five lab shelves that is fifteen sentences describing five pages, and fourteen of the fifteen were different from one another. The only two that matched had been rewritten together, in one sitting, the day before.

None of them was false. Each was written by somebody solving one audience’s problem, and each was accurate on the day it was written. The problem is what happens next: when a page changes, whoever updates it fixes the sentence they can see. Nothing in the site read two of the three at once, so nothing could notice the others had been left behind.

We have published this exact failure before. In August we corrected a case where a single lab note described itself two different ways, and that fix reached the notes and stopped there. It never reached the shelves the notes sit on. You can read that one at We Described the Same Page Two Different Ways.

From today each lab shelf has one description, written once, and the reader’s page, the search result and the structured data all read it from the same place. Change it once and it changes everywhere. Where a page needs different words for a good reason, that difference now has to be written down as a decision, so nobody has to guess later whether two sentences drifted apart or were meant to differ.

Having one description each is not the same as having a true one. Read against the records underneath them, two of those five sentences were describing a different page: the System Updates shelf promised methodology changes and holds capabilities, and the Decision Logs shelf promised thresholds and holds rules.

The second change is for machines, and it is the larger one. Source Corrections is our public record of things we published that turned out to be wrong. To a person it has always been a list. To a machine it was a title and one sentence, because the page never listed its own contents in the structured data underneath it. Our record of our own mistakes was the one page where being readable by machines matters most, and it was invisible.

It now publishes its members: each entry’s title, its address and its position in the list. So does every other lab shelf that was missing one. We built this three times before on individual pages, once per page, each time for the page in front of us, and each time the next page inherited nothing. This time the rule asks whether a page has already listed itself rather than which page it is, so there is no fourth page left to forget.

Two things this does not fix. The archive still does not tell a reader how many corrections it holds, even though it now tells a machine. And the robots file still describes these shelves in its own short lines, which we have deliberately left alone and written down as a decision rather than a gap.