I found a number on my own homepage that used to be true. It sat in a terminal-style footer block as "LCP ... 2.3s OK", and a receipts line said "homepage LCP 7.0s to 2.3s". Both were typed in by hand after an earlier round of fixes. Nothing on the page measured them, so nothing noticed when they went stale.
That's the quiet problem with hardcoded stats. You write the number once, the page stops checking it, and it keeps looking confident. No test fails. No alert fires. The only person who notices is someone who opens the page with the browser's dev tools open, or who runs PageSpeed and compares.
Then I ran PageSpeed on the live homepage and got Performance 96, Accessibility 94, Best Practices 96, and SEO 100. Those numbers are real, and the 2.3s was not what the page should have been showing next to them. I replaced the old line with the current scores, but the lesson I took is simple: if a claim on a page can drift, it should come from somewhere that can be checked.
The other flag was "Links are not crawlable." The culprit was a drawer link on my site. JavaScript built the anchor with a target and a rel attribute, but the href only got set on hover, when the drawer opened. To a crawler, and to PageSpeed, that element is an anchor with no destination. Humans who hovered could follow it. Everyone else saw a dead link.
The fix is a default href on the anchor, with the hover logic still free to override it per item. I'm keeping the hover behavior because the drawer should still change where it points. The real lesson is that a link which only exists on hover is only a link for people who hover.
A third problem turned up while I was checking this. A line in the LED bay said "last deploy 33d ago," while the deploy strip showed a deploy the day before. Two parts of the same page disagreed, and I still haven't traced where the stale text comes from. That's the usual pattern: counters and "last updated" lines rot first, because nothing ties them to the thing they describe.
Here's what I'm changing in my own workflow. Any stat on a public page gets a date and a device or source next to it, like "measured Nov 21, mobile." Anything I can compute, like a deploy age or a count of shipped items, should be computed at render time rather than typed in. And I'm treating a PageSpeed flag as a clue to check the markup the crawler actually sees, not just the markup I think I wrote.
If you're running PageSpeed on your own site, try this: view the page source, not the DOM, and look for any number that doesn't have a date next to it. That's the one I'd fix first.