Our Correction Records Now Open With What We Got Wrong

When we publish something wrong and then fix it, we write the whole thing down and leave it up. Those records live in our lab, alongside the logs of rules we changed and the notes on what the site gained. Until this week they were the last pages in the lab still built the old way, and the old way buried the point.

You arrived at a headline. Under it sat a bordered card holding four rows of filing information, and the first of those rows was the sentence that says what we got wrong. It ran to about a hundred and eighty characters and was set at the size of the filing labels beside it, less than half the height of the headline above and smaller than the article below. On a phone that card filled most of the first screen, and the writing began near the bottom edge.

So the one sentence a reader needs in order to know whether a page concerns them was dressed as bookkeeping and sized to match.

That sentence is now the opening line. It sits directly under the headline, set large, in the same face, exactly where every other kind of lab record has carried its opening sentence since we rebuilt them last year. The judgment underneath it moved with it: every correction we publish carries a plain statement of who it touched, either that a reader could have relied on the wrong thing or that it never left our own machinery, and that statement now reads as the line under the claim instead of as row two of a table.

The date the record last moved became a small line above the headline. The filing details went to a block further down, next to the tool that hands the whole record over to anyone who wants to cite it, which is where that kind of information was always going.

One thing on that page is new rather than moved, and it is the part worth knowing about. When we correct something, we usually go back to the page that carried the wrong claim and leave a dated notice on it, so a reader who met the mistake there meets the correction too. The first of those notices went up in January. Nowhere on the correction itself did we say which pages got one. Now we do: the record names them, and each one is a link, so you can go and read the notice where it sits. Twenty-one of our fifty-one corrections placed a notice somewhere; the widest of them placed nine.

The last change is the footer. These pages ended with a line saying we took no payment from any brand mentioned and that our recommendations and rejections are independent. On a correction about our own machine files, no brand is mentioned, nothing is recommended and nothing is rejected. It was a product-review footer under a page that reviews nothing, and it quietly suggested the page was about a product. These records now say what they actually are, and the payment line stays only on the eight corrections that touch a product page, where it means something.

Two things did not change, and we would rather say so than let you find out. The system updates shelf, this one, keeps the old layout for now. The sentence we would have promoted is written as a headline on some of those records and as a claim on others, and setting a headline in large type under another headline gives a page two of them. It is a writing problem rather than a design one and we are not going to paper over it with a rule that guesses which is which.

The second thing lasted about an hour. On the corrections that placed no notice, the block near the foot of the page came up holding a single row naming the shelf the record is filed on, which the trail at the top of the page had already named and which the link to the next record names again. One line of furniture, correct in every detail, doing nothing. We read the built page, saw it, and took the row out the same afternoon. Where that block now has nothing to say it does not appear, and the tool for citing the record sits on its own, which is what it did before.