The Infobox Doesn't Know About the Prose
The pitch for wiki reliability has always been the same: enough eyes on a page, enough time, and errors get caught. Vandalism gets reverted in minutes. A wrong date gets fixed by whoever notices next. The page converges on accuracy because editing is cheap and continuous.
That pitch has a hidden assumption baked into it: that there's one surface being edited, and the eyes watching it are watching the same thing. A lot of pages don't work that way anymore.
Take a small municipality page, the kind with one real sentence of prose and everything else scaffolding. The body text cites a fourteen-year-old national census figure. The infobox at the top of the same page shows a different number, newer, pulled automatically from a structured database that gets updated on its own schedule by a process that has never read the prose and never will. Both numbers sit on the same page, a few hundred pixels apart, off by more than a hundred people. There's a maintenance tag flagging the population figure as needing an update. It's been sitting there for months. Nobody's reconciled anything.
This isn't vandalism, and it isn't neglect in the usual sense either. Neglect implies someone was supposed to look and didn't. Here, two different someones are looking, constantly, and neither one considers the other's territory part of the job. The bot's task is "keep the infobox current against the database." It executes that task perfectly, on schedule, forever. The human editors watching the page are reading prose, not infoboxes, because prose is where their attention naturally lands and infoboxes read as settled, machine-generated, not the kind of thing you'd think to fact-check against the paragraph next to it.
So the page has two write surfaces running on two clocks, and the whole self-correcting mechanism only operates within each surface separately. Vandalize the prose, someone reverts it. Feed the bot a bad number, the next sync corrects it. But let the two drift apart from each other, quietly, correctly, each one accurate by its own lights, and nothing on the page is built to notice. There's no diff between infobox and prose. No bot compares its own output to the sentence sitting six inches away and flags a mismatch. The reconciliation step that the whole reliability story depends on just isn't a step anyone assigned to anybody.
It's not limited to population counts on stub pages. Any page that stitches a live data feed to hand-written text has the same seam. A product page where the spec table is pulled from a vendor API and the review prose above it was written eighteen months ago. A company profile where the "employees" field updates from a filings scrape while the founding story paragraph hasn't been touched since it was written. A sports roster infobox that updates the day a trade happens while the biographical paragraph still describes the player on his old team. The infobox is telling the truth. The prose was telling the truth when someone wrote it. Read together, they tell you the page hasn't been looked at as a whole in a while, and nothing about the page format will ever say so on its own.
What makes this different from ordinary staleness is that it looks like currency. An infobox with a live number reads as more trustworthy than static prose, not less, because it's obviously being maintained by something. That's exactly backward. The freshness of one half is doing nothing to vouch for the other half, and the page gives you no way to tell the difference between "this whole entry is current" and "one field on this entry is current and the rest hasn't moved in a decade."
The fix isn't complicated in theory, some kind of diff check between structured field and cited prose value, flagged the moment they diverge. Nobody's built it in, because the two systems were never designed as one system. They were bolted together, each doing its job well, and the seam between them is exactly where the page stops knowing what it's saying about itself.