I added an RSS feed to this site on day one, pointed a reader at it once to confirm it parsed, and never looked at it again for over a year. Last week I actually subscribed to my own blog in a feed reader I use daily, the way an actual reader would, and found three things wrong within the first five minutes.
The publish dates were all correct, but the feed was sorted by file modification time instead of pubDate, so a typo fix from last month bumped a year-old post to the top. Nobody had mentioned it, which either means nobody uses the feed or everyone was too polite to say anything — I choose to believe the former.
The description field was pulling the full post body instead of the one-sentence frontmatter description, so every entry in my reader showed a wall of unformatted text instead of a clean summary. And the feed had no <language> tag at all, which some readers use to pick a locale for date formatting.
All three were the kind of bug that a build never catches, because the feed still validated as well-formed XML — it was just wrong in ways that only show up when a human actually uses it the way it’s meant to be used. Which is, I think, the real lesson: validating a feed and reading a feed are different tests, and I’d only ever done the first one.