Something we published turned out to be wrong, and here is the correction, dated and kept in public. Makers correct us, and so do our own checks.
A fact on a page, which a reader could have relied on
Three Lab records whose subject is our own pipeline sat under Deep Dives, the shelf for published research. Two are now System Updates, one a Decision Log. No URL changed.
Three records left Deep Dives; the shelf's published count fell from 66 to 63
A fact on a page, which a reader could have relied on
We published that shorts have a dedicated sitemap while the address returned 404. The cause was a rewrite flush hooked to theme switch, an event our releases never fire.
Correction - the shorts sitemap we announced returned 404 for a day; the route now registers on every theme release
Our own machinery, which no reader would have acted on
Six product pages printed a verdict heading our own template had cut mid-sentence, because the cut lands at the first comma and those sentences opened with a list.
Six product pages printed a heading our template had cut mid-sentence
A fact on a page, which a reader could have relied on
Recommendation links on twenty seven product pages pointing at four products that are no longer on the site
Twenty eight recommendation links pointed at products that no longer exist. Removed, reasons kept, whole site re-swept.
A fact on a page, which a reader could have relied on
The copy-ready reference under every lab note joined its parts with a full stop, so a title already ending in one published two; and passed notes named their subject to readers and to no machine.
The reference takes its separator from what the part before it ends with, and a passed note now names the product it tested.
Our own machinery, which no reader would have acted on
Nine research pages printed a paper's web address with its prefix repeated, and a document describing our data to AI systems named the wrong job for one of our fields.
A doubled address prefix is trimmed at every reader, and a published field description that named the wrong consumer is corrected.
A fact on a page, which a reader could have relied on
The Decision Logs shelf told readers it was about the thresholds behind our methodology. One of its sixteen records is about a threshold. All sixteen are a rule of ours changing.
The shelf now describes itself as a record of rules that changed, which is what eight other surfaces here already called it
A fact on a page, which a reader could have relied on
The sentence under this shelf's heading said it held changes to our methodology and scoring system. Our own routing rule sends those to Decision Logs, and no record here is one.
The shelf now describes the record it actually holds, and the page itself was rebuilt in the same release
A fact on a page, which a reader could have relied on
Our search listing for the Tensira throw rejection gave its final score as 7.6. The score we recorded when we turned it down was 7.80.
A rejection's search listing published a lower score than the one we recorded
Our own machinery, which no reader would have acted on
Two machine-readable documents described each score band as a top percentile of our catalog. Counted against every published score, three of the four were wrong.
The rarity claims are removed, and the bands now come from one definition instead of three copies