A correction changes a paper. A retraction changes its status.

A published paper can need a correction without becoming an unreliable paper. It can also become unreliable in a way that calls for retraction. The labels matter because they tell a reader what kind of problem the publisher is addressing.

BMJ’s Author Hub policy, updated June 2024, describes corrections to published work as linked correction notices for factual errors. It also says that minor typographical preferences that do not affect scientific understanding may be rejected. Its retraction policy covers cases such as unreliable data or findings, plagiarism, duplicate publication, and unethical research. During an investigation, BMJ may publish an expression of concern. These are BMJ’s policies, not universal rules for every journal.

Consider a fully fictional paper. Its methods section says the sample contained 120 records, but the verified count is 121. The analysis and conclusion use the correct count, and the journal publishes a linked correction. That notice changes a factual detail while leaving the paper’s central finding intact.

Now change the fictional case. The paper’s central result depends on a dataset later found to be unreliable, and the conclusion cannot be supported by the available evidence. A retraction notice addresses the status of the publication. It does not require the reader to pretend the paper never existed, and it does not by itself prove that the authors acted dishonestly. An honest error can produce an unreliable finding.

BMJ says the original text usually remains accessible with a retraction notice or watermark, with rare exceptions involving legal removal. That means an old abstract or cached copy should not be read alone. Follow the publisher’s notice and inspect what it says changed, why the action was taken, and whether the concern affects a number, a method, or the paper’s central finding.

My practical rule is to treat the label as the beginning of reading, not the end. A correction asks you to locate the changed fact. A retraction asks you to reconsider the reliability of the published work. An expression of concern tells you that an investigation is still underway. None of those labels, by itself, supplies every detail a reader needs, and none should be stretched into a universal verdict about the authors.