Every claim Cento makes about a citation is one you can check yourself, in a click. Not because we ask for your trust — because the source is right there to open.
Cento only cites papers it actually retrieved from the open literature. Every citation carries a working link to the source — the abstract, the DOI, the record. If a reference can't be opened, it isn't a reference; it's flagged.
A randomized trial and a single case report are not interchangeable evidence, and Cento never lets them look that way. Each source carries its level of evidence inline, so a weak source can't quietly stand in for a strong one in your argument.
A real, well-graded source still has to actually support the sentence it's attached to. Cento scores the alignment between your claim and its source and watches it as you revise. When a sentence outruns its evidence, it flags as Drifted — before a reviewer makes the same observation.
We could put a single accuracy percentage at the top of this page. We don't — because one number, averaged across every kind of claim and every corner of the literature, would tell you almost nothing about your paper.
The test that matters is the one you run yourself: open the sources, read them, and check them against your sentences. Cento is built to make exactly that test fast — every claim one click from the evidence behind it. As clinician users come on board, their published work, not our marketing, will be the proof.
No source, no sentence. The claim is marked [UNCITED] instead of dressed up.
Citations bind to retrieved candidates by ID, or they don't exist at all.
Cento is a co-writer. You remain the scientist who decides what the evidence says.
A weak source is graded weak; a drifted claim is flagged, not smoothed over.