August 23, 2026
We Called Them Thresholds and They Are Rules
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.
Our Decision Logs shelf collects the rules we changed. Its own opening sentence said something narrower: that it held the judgment calls behind our methodology, and why we set thresholds where we did.
We counted. One of the sixteen records on that shelf is about a threshold, and it is the only one of them that contains a number at all. Three touch the scoring system. All sixteen are a rule or a standard of ours changing.
So the second sentence described about six percent of the page it introduced, and a reader who arrived for thresholds found rules.
The odd part is that the rest of the site had it right the whole time. A record’s own page labels the field carrying its claim “the rule that changed”. Three pages send readers to this shelf under the words “every rule change”. Both files we publish for machines describe it as a rule, threshold or standard of ours changing. Eight surfaces, one wording, and the page a reader actually lands on was the exception.
The sentence now says what the records are, in the wording the machine files already used, so a reader and a crawler meet one description instead of two.
Why nobody caught it sooner is the part worth keeping. The day before, we gave all five lab shelves one description each, held in one place, so the wording could not drift apart between the page, the search result and the structured data. That was the right fix and it was done well. What it never did was ask whether the surviving sentence was true. Unifying where a claim lives and checking whether the claim is correct are two different jobs, and the first finishes looking like the second.
This is the second shelf found this way in two days. The first was System Updates, earlier the same day, and the note we wrote then said the same thing could be sitting next door. It was.
The page was rebuilt in the same release. It now sorts its records into four groups by what each rule governs, which is something the shelf could not do before, and that change is written up on its own.
One field decides those groups, and it now does a second job: every record closes by naming what its rule constrains and sending you where it operates.