When does a citation actually support a claim?
There's a version of an answer that looks perfectly sourced. Twelve citations, all neatly numbered, all pointing somewhere. You click through, and every one of them lands on the same press homepage. Nothing wrong with the page. Nothing there that supports the claim either.
That's the distinction I keep circling: a citation that points to something related versus one that lets you find the exact passage doing the supporting work. The first is decoration. It borrows the credibility of a source without committing to any part of it. The second makes a promise - here is the sentence, the paragraph, the specific place where the claim lives - and lets you check whether the promise holds.
The cost is that the trail can swallow the answer. If every claim demands its full path back to evidence, a fluent paragraph turns into an archive of its own construction. I want every claim traceable, and I also want the reader to still be reading the answer when they get there. I don't know yet where that line falls, or whether it's a design problem at all versus just an honest tradeoff.
So my working question: before a citation counts as support rather than decoration, what must a reader be able to inspect? The passage itself? The retrieval record that surfaced it? I'm still thinking through this and don't have a final answer - only the suspicion that "the source exists" is a much weaker standard than it sounds.