A set of figures got shuffled somewhere between the proof and the printed page. That is the heart of a correction notice published on July 30, 2026, in BMC Veterinary Research, filed by the fifteen authors of a study on predicting the body weight of Blackbelly sheep from biometric measurements using two machine learning methods, XGBoost and Random Forest.
The correction is short and unglamorous, and it is worth reading anyway. Corrections are the part of science publishing that most readers never see, the quiet mechanism by which the record gets tidied after the fact.
Here is what the authors report. After the original article appeared, they went back and read the final published version carefully. Several corrections that had been marked in the proof file, the near-final version an author reviews before printing, and which the corresponding author had gone through, simply were not there in what got published. The proof said one thing. The article said another.
The second problem is the more consequential one. The order of the figures, the authors write, "appears to have been mixed during production," and some figure captions no longer match the figures they sit beneath. In a paper built on comparing how two algorithms perform, this matters a great deal. The authors say so directly: because figures are essential for accurately interpreting the results, the mismatch could confuse readers.
The notice ends with a single sentence of resolution. The original article has been corrected.
What the correction does and does not tell us
It is worth being precise about the scope here, because a correction notice is a narrow document and it is easy to read more into it than it contains.
The notice does not say the study's findings were wrong. It does not say the models misperformed, or that any number in the analysis needs revising, or that the underlying data had problems. Nothing in it points to an error in the science itself. The problems described are production problems: text changes that went missing, figures that ended up in the wrong sequence, captions attached to the wrong images.
Nor does the notice tell us what the original study actually found. It does not report how accurately XGBoost or Random Forest predicted a sheep's weight from measurements like body length or chest girth, nor how many animals the researchers measured, nor which of the two algorithms came out ahead. Those details live in the original article, published earlier in 2026 in the same journal, and the correction simply points readers there.
What the notice does establish is that anyone who read, cited, or tried to reproduce the figures in the original version between publication and correction may have been working from a scrambled visual record. If you looked at a plot and read the caption beside it, the two may not have belonged together.
The authors themselves caught it. Not a reader, not a reviewer, not an automated check: the people who wrote the paper, rereading their own work after it went out into the world.
Why it matters
Blackbelly sheep are raised across the tropics, and knowing an animal's weight matters for practical reasons: dosing medication correctly, deciding when to sell, tracking whether a flock is thriving. A livestock scale is expensive and awkward to move. A tape measure is neither. That is the appeal of predicting weight from a few body measurements, and it is why researchers keep testing whether modern statistical methods can do the job well enough to trust in a field.
Which makes the figures in such a paper more than decoration. When the point of a study is to compare how two prediction methods perform, the plots showing predicted weight against actual weight are the evidence. A reader who cannot match a plot to its caption cannot check the claim. The authors evidently understood this, which is why the mismatched captions, rather than the missing text corrections, get the more urgent framing in their notice.
There is a broader point about how publishing works. Between a manuscript an author approves and the article a reader downloads sits a production process involving typesetting, file handling, and layout, and things can go wrong in there without anyone intending them to. Authors review proofs precisely to catch this. In this case the proof review happened, and the corrections still did not make it through.
The reassuring part is that the system did eventually work. The authors read their published paper, noticed the discrepancies, told the journal, and the journal issued a public notice and fixed the original. A reader arriving now gets the corrected version. A reader who saved a PDF in the weeks before might not, which is a small argument for checking whether an article you are relying on has a correction attached. Most do not. Some do, and this one now says so on the record.