IOP Publishing now runs misconduct cases through six research-integrity officers

The two headline counts in that survey do not share a denominator. The 782 is a tally of duplication instances ^[3]; the 196 is a tally of papers ^[4]. Set the 196 against the more than 20,000 papers screened and you get under 1 per cent, and a little under that, since the sample was described as more than 20,000 ^[1]. Dividing 782 by 20,000 gives about 3.9 per 100 papers, though that counts events and not documents ^[3].

Both numbers describe a failure mode that leaves a mark in a figure. Duplication and alteration are visible in the published record; a fabricated number that was never plotted is not. The papers surveyed appeared between 1995 and 2014 ^[2], a twenty-year window ^[4] that closed before the AI and machine-learning tools that Physics World says let bad actors rapidly increase output and target journals with weaker checks ^[16].

Bik had been sending reports on plagiarism and image duplication to journal editors anonymously, prompted by finding that someone had plagiarized her own work ^[5]. Those reports each concerned one paper; the survey counted across 40 journals ^[2]. She went on to advise and mentor the amateur sleuths who followed ^[6], and in 2024 the Einstein Foundation Award cited her for "identifying misconduct and potential fraud in scientific publications, highlighting science's problems policing itself" ^[7].

Inside publishers, the work now has a job title. IOP Publishing, which publishes Physics World, employs a team of six research-integrity officers handling misconduct cases ^[9]. Michael Banks, writing for Physics World, does not say how many cases that team handles or what screening tools it uses. He describes misconduct as a far bigger problem than 10 or 20 years ago, when fabrication, plagiarism and self-plagiarism cases in physics were notorious for their rarity ^[10].

The old cases were individuals. An independent committee found in 2002 that the Bell Labs physicist Hendrik Schon had fabricated data and falsified reports in 16 of 24 papers published between 1998 and 2001 ^[11], two thirds of that output ^[2]. The University of Rochester, where Ranga Dias worked after he claimed superconductors operating at high pressures and ambient temperatures, concluded that he had committed misconduct including data fabrication ^[12]. Papermills are a different target: businesses that sell authorships, citations and data, and produce full papers ^[14], marketed to students with promises of guaranteed acceptance and citations ^[15].

Physics World reports that papermills have evolved quickly to evade publisher checks and the peer-review process ^[15]. Against that, six officers is a fixed capacity. In my view, staffed screening will keep catching the duplicated image, which is cheap to spot, and will keep missing the generated paper that ships clean figures. A published detection rate per thousand submissions, measured against known papermill output, would settle that.