We Told You What Kind of Page You Were On

When we test a product we publish the reasoning, whether it cleared our bar or not. Near the top of every one of those pages sat a short line naming what was tested: the product, the maker, and the kind of thing it is. That third item was wrong on all 585 of them.

Instead of the product’s category it printed our own filing label. A fountain pen read as Stationery in our records and as Lab Results on the page. Lab Results is the shelf we keep these write-ups on. It is not a kind of object, it is not something anyone was searching for, and it told a reader who had just arrived from a search engine what sort of page she had landed on rather than what we had put on the bench.

The odd part is that we already held the right answer, on both kinds of page. Where a product did not clear the bar, its category is stored on the write-up itself. Where it did, the write-up is tied to the product’s own page and the category is stored there. Either way the section further down the page was already reading that value, to show you other things in the same category that we do recommend. The one line whose whole job was to answer the question read a different field.

The line now reads the product’s own category, and it has moved down into the record block at the foot of the page. 584 of the 585 resolve to a real category, checked one by one against the records they read from before the change shipped. The one that does not is a write-up with no product attached to it, and it now shows no category at all rather than the wrong one.

It was not a typo in a sentence, which is why nothing we own caught it. Our checks ask whether a published claim is true, whether a number still matches its source, and whether a page says what our records say. This page said something true, from a real field, correctly spelled. It was simply the wrong field, and no instrument we had was looking at that.

Editor’s note, added 20 August 2026: this entry first published on 19 August and said the line was wrong on 198 pages. 198 is the number of products that did not clear the bar, which is where we found it. The same wrong value was printing on the 387 write-ups that passed, through a different part of the same page, and it was corrected there the following day. The count is 585, and it is corrected here rather than in a second entry about this one.