Four Documents, Two Wrong Years, One Slot’s Research. The Oldest Was Thirteen Years Old and Its Arithmetic Was Perfect.
A single slot’s research returned four documents from the wrong period, across two distinct wrong years — three from 2025 and one from 2013 — and every one of them returned a clean HTTP 200 and rendered in full as prose.
The thirteen-year-old one is the lesson. Searching for Japan’s October trade balance returned an article headlining a deficit of 1.090 trillion yen that passed every test a pipeline runs: status 200, text/html, 32,572 characters, no redirect; roughly twenty populated fields with nothing missing; a plausible magnitude, 1.14 times the consensus actually published for this release; a day-of-week that matched, because 18 November 2026 is a Wednesday and the article says the data came out on a Wednesday; page chrome dated 11 to 17 November 2026 in the sidebar, slide-show titles naming November 2026 and a 2026 copyright; and, worst of all, perfect internal arithmetic — exports of 6.104 trillion yen and imports of 7.195 trillion give 1.091 against a stated 1.090, a one-billion-yen gap that is rounding. The figures are October 2013’s. The byline under the headline reads Published: 11/19/2013 8:03 PM ET, and the body names no year at all, not in the headline, not in the prose, not in the meta description. The real print, carried by three publishers, was 476.5 billion yen, so the stale document is 2.288 times the actual number. The failure was not a missing date field: a reader found that byline, reported it, and then argued it away as a template error because the surrounding items were current — which inverts the only rule that works, that page chrome is authoritative about the publisher’s present and silent about the document’s past. The weekday check is a one-in-seven test and this pass drew both outcomes: it correctly rejected an undated option-expiry page titled for Monday the 17th and Tuesday the 18th, because in 2026 those are a Tuesday and a Wednesday, and it waved the trade article through because the byline is stated in United States Eastern and the article’s weekday in Tokyo. On none of the three undated documents is the year a field at all; the day and month sit where a filter reads and the year does not. The checks that would have worked are free: the publication identifier in the URL, 2226882 against sidebar identifiers near 3,713,000, implying 1,486,118 documents and roughly thirteen years; the content type against the extension; and the page’s own meta tag telling news crawlers not to index it. Arithmetic consistency is the one that cannot help, because self-consistency tests hygiene rather than vintage.