Every Sentence Traced And The Page Still Said More

Every claim on a product page here is tied to a source we can walk back to, and that walk is written down so you do not have to take our word for it. The machinery works. This is about what goes wrong one layer above it.

Five defects were found by people reading pages. Not by any instrument. In each one the sentence traced perfectly: the source existed, the quote was word for word, the walk passed. What failed sat on top of the trace. An arithmetic nobody had counted. A circumstance borrowed from the claim next to it. A hedge dropped while a long sentence was compressed. A superlative over a set of one. A characterising word that entered in our own summary rather than in the source. Every grounding gate we own passes all five by construction, which is why no instrument was ever going to raise them.

The clearest case is a gate that was already right. It has said since its first version that it checks for editorial language stronger than the source’s own words. That reads as a rule about our language about a source, and a reported conclusion does not look like that. It looks like narration. So four instances of a speaker arriving on the page surer than he left it went straight past it. The repair was to name the case, not to add a rule. A correct rule nobody applies to the case it covers is not a safety net.

The standards that changed govern what a page may assert beyond what it can trace, and they now reach how a page is arranged. A surface that cuts a heading out of somebody’s prose, that fills one repeated group from several unrelated fields, that shows a quantity before anything establishes its scale, that hides one value because another is believed to contain it, or that shows a machine more than it shows a reader, is making an assertion in every one of those cases and writing no words in any of them. No word scan reaches it, and every writing gate reads it as furniture.

What that found on the live site, and neither is repaired today. One heading on our product pages is cut out of a sentence the writer wrote: the cut fires on 160 of 384 records, and on 159 of those it lands on a comma, so the template supplies a full stop the writer never wrote and renders the fragment as the room’s heading. And on the review cards: 273 records carry rated voices, 228 of them publish only four and five star ones, and on 105 of those the page’s own prose says opinion is split. Both are owned, both are open, and we would rather publish the measurement than the intention.

Underneath all of it sits the same registration standard, written the same month: before a field can carry anything, it has to be established what its emptiness means to everyone reading it, what shapes it may legally take, and what a template may say about it. A template may state what a field contains. It may never state what those contents mean.