This week, Nature publishes retractions of two high-profile papers that claimed a major advance in the field of stem cells (see page 112). Between them, the two papers seemed to demonstrate that a physical perturbation could do what had previously been achieved only by genetic manipulation: transform adult cells into pluripotent stem cells able to differentiate into almost any other cell type. The acronym STAP (stimulus-triggered acquisition of pluri­potency) became instantly famous.

Soon after the papers were published on 30 January, cracks appeared in the edifice. As various media outlets including Nature’s independent news team reported, errors were found in the figures, parts of the methods descriptions were found to be plagiarized and early attempts to replicate the work failed.

The problems that initially emerged did not fundamentally undermine the papers’ conclusions. Moreover, replication of such work is not necessarily straightforward or quick, and the ability to use some techniques can be very sensitive to aspects of the experimental protocol.

Nevertheless, the RIKEN research centre in Japan, one of the institutions in which most of the work was carried out, promptly organized an inquiry and found inadequacies in data management, record-keeping and oversight (see go.nature.com/2vrjxs). One author was found guilty of misconduct — a charge that RIKEN reaffirmed following an appeal (see go.nature.com/tnxuhy).

We at Nature have considered what lessons we can derive from these flaws. When figures often involve many panels, panels duplicated between figures may, in practice, be impossible for journals to police routinely without disproportionate editorial effort. By contrast, image manipulation is easier to detect. Our policies have always discouraged inappropriate manipulation. However, our approach to policing it was never to do more than to check a small proportion of accepted papers. We are now reviewing our practices to increase such checking greatly, and we will announce our policies when the review is completed.

But only since the RIKEN investigation has it become clear that data that were an essential part of the authors’ claims had been misrepresented. Figures that were described as representing different cells and different embryos were in fact describing the same cells and the same embryos.

All co-authors of both papers have finally concluded that they cannot stand behind the papers, and have decided to retract them.

The papers themselves have now been clearly watermarked to highlight their retracted status, but will remain hosted on Nature’s website, as is consistent with our retraction policy. (In our opinion, to take down retracted papers from journal websites amounts to an attempt to rewrite history, and makes life needlessly difficult for those wishing to learn from such episodes.)

We at Nature have considered what lessons we can derive from these flaws.

We at Nature have examined the reports about the two papers from our referees and our own editorial records. Before publishing, we had checked that the results had been independently replicated in the laboratories of the co-authors, and we regret that we did not capture the authors’ assurances in the author-contributions statements.

We have concluded that we and the referees could not have detected the problems that fatally undermined the papers. The referees’ rigorous reports quite rightly took on trust what was presented in the papers.

For more than two years, Nature has been publishing articles that highlight the problems that can arise in the analysis and publication of the results of biomedical research. We have also launched several initiatives to improve our own rigour. For a collection of relevant content, see go.nature.com/huhbyr.

Underlying these issues, often, is sloppiness, whether in the handling of data, in their analysis, or in the inadequate keeping of laboratory notes. As a result, the conclusions of such papers can seem misleadingly robust. Another contributory factor lies in selection bias behind the data presented, whether implicit because the experiment was not randomized or blinded, or explicit in the deliberate selection of data that, usually with honest good intentions, are judged to be representative. (This is not to say that randomizing and blinding is always required, but more is needed than currently occurs.)

A manifestation of these problems has been a growth in the number of corrections reported in journals in recent years. It is hoped that the extension of our methods sections, the addition of a checklist intended to improve the standards of reporting, and our use of statistical advisers will reduce these problems in Nature.

In short: although editors and referees could not have detected the fatal faults in this work, the episode has further highlighted flaws in Nature’s procedures and in the procedures of institutions that publish with us. We — research funders, research practitioners, institutions and journals — need to put quality assurance and laboratory professionalism ever higher on our agendas, to ensure that the money entrusted by governments is not squandered, and that citizens’ trust in science is not betrayed.