Most "AI finds relevant papers for you" pitches are vague on purpose, because the claim underneath is usually just a better search box. Here's a more specific, more checkable one: suppose a scheduled run surfaces a preprint that cuts against a claim in one of your drafts. When draft comparison is available, Folio can label that finding as challenging the draft and attach a short reason. It does not claim to identify the exact sentence; it takes you back to the document so you can inspect the argument yourself.
That's the new part. Discovery Agents already watch a field on a schedule and rank what shows up. For eligible plans, each scheduled run can compare up to three of the strongest finding-to-draft matches and form a judgement: does this paper support your argument, challenge it, or is it just related background? The result is a verdict with a one-line note explaining the reasoning, shown with the finding in the agent view and briefing email.
Where you actually see it
Open a finding's card in the agent view and, if it bears on something you're writing, you'll see a colored line — supports, challenges, or related to — linking to the draft in question. Click it and Folio opens the document. It does not pretend to know the exact paragraph.
The same judgement rides along in your briefing email, under a short "What this means for your drafts" section: the paper on one side, your draft on the other, and the note in between. The draft title links back to the editor. The paper title links to its source when the result includes a safe source URL.
A concrete flow: your agent runs its weekly pass and surfaces a paper reporting a null result on a mechanism your draft treats as established. The briefing email arrives with a red "Challenges your draft" line under that paper, pointing to the draft. You open the document, find the relevant claim, and decide — cite it, hedge it, or note why your case still holds. Either way, you made that call with the contradicting evidence in front of you instead of a reviewer making it for you six months later.
Why this is the useful version of "AI finds papers"
Citation drift is quiet. A paper you built an early paragraph on gets contradicted, corrected, or superseded, and nothing tells you — the paragraph just sits there, increasingly wrong, until a reviewer or your own re-read catches it. Catching that during drafting is a different experience than catching it after submission. It's a smaller fix, made earlier, with more choices still open to you.
This is also why we've been resistant to shipping a generic "relevance score." A number doesn't tell you what to do next. A verdict with a reason and a link back to the document gives you something concrete to examine.
The judgement is cached from the scheduled run, so it's there the moment you open the finding; asking the agent directly always gets you a fresh check against the draft as it stands right now.
Discovery Agents watch your field and now watch your drafts against it too. See how they work.