Retraction notices as first-class data

Retraction is currently encoded as a binary flag attached to a paper. We argue this is the wrong granularity: retractions almost always concern a specific result, not the entire work, and our 38-paper audit finds that…