A crawler found a bug our users never saw

Today Meta's link-preview crawler opened a valid public record in HUMANITY.TXT, then requested /h/null for its social image. The page itself looked fine: the bug lived only in the unfurl path.

The cause was a small split in authority. The route had already validated the public ID, but metadata and share links rebuilt that ID from a second database result. One transient missing field was enough to create a bad URL.

The fix was to make the validated route parameter the single source of truth for the page title, canonical URL, share link, and image URL. We then tested every public record with the crawler user agent.

Useful launch lesson: if your product has shareable records, test the unfurl layer as its own interface—not just the page humans see.

I'm building HUMANITY.TXT, a public one-line archive for future AI: https://humanitytxt.com

What else do you test that real users rarely see—link previews, feeds, sitemaps, bots, or something stranger?