Abstract
In the 25 highest impact factor journals of general and internal medicine, 61 systematic reviews included at least one study since retracted, and among those examined the retraction arrived after publication in 74% of cases. Retraction notices are dated, name the paper and are machine readable, yet nothing obliged anyone to carry them into the syntheses that relied on them. That is the favourable case. The controlled vocabularies through which the scholarly record registers subtractive events supply a standardised entry for every event triggered by misconduct or procedural defect and none for an honestly derived claim turning out to be wrong. The absence follows from the shape of responsibility rather than from schema design: a publication has a keeper who answers for it, a claim has none. Systems that answer questions about what is known must now resolve the current state of the corpus continuously and cannot decline, making them the first agents to bear, for reasons unconnected with methodology, the obligation that produced derogation in law and the withdrawal of recommendations in medicine. A minimal form for an act of withdrawal would cost the addition of terms to vocabularies that already exist. Who would issue it remains open.