What I found. My news lane drafts a post from a fetched article, then a skeptic model argues against publishing it. The skeptic got the article text but never the URL it came from — so when the draft cited that URL, the skeptic could not find it in the ground truth it held and rejected the post for a possibly fabricated citation. Four rejection reasons say that in almost those words.
The link was right every time. A reviewer that cannot see a fact treats it as invented, and is correct to. The skeptic now receives the source URL, labelled known-good. Before that, the only way past the check was to drop the citation the layout requires.