Do not begin reviewing a translated research paper by polishing its introduction. First find the result you are most likely to quote or act on. If its claim, numbers, or references changed, elegant sentences elsewhere will not make the file dependable.
Keep the original and translated files open together. Work in a separate review note so you can distinguish a factual correction from a stylistic preference.
Read the abstract and conclusion as a pair
Underline the central claim in the original abstract, then find its counterpart in the translated abstract. Identify the method, subject, comparison, and stated outcome. Repeat that exercise in the conclusion.
Now compare the two translated passages. Does the conclusion claim more than the abstract? Is that difference also present in the original, or did translation introduce it? A qualifier such as “in this sample” may be essential rather than optional background.
Pay attention to degrees of confidence. “Suggests,” “is consistent with,” and “demonstrates” are not interchangeable. Do not strengthen a cautious statement merely because a stronger verb reads more cleanly.
If an apparent contradiction is already in the source, record it as a source question. Editing a translation is not permission to silently repair the authors' argument.
Verify numbers with units, conditions, and negatives
Choose important values from the findings and compare them directly. Include sample sizes, percentages, ranges, thresholds, and any number that changes an interpretation. Search for digits to locate candidates, but do not assume that a search catches written-out numbers or a lost symbol.
Check the text attached to each value. “No more than,” “at least,” “not significant,” and “except” can reverse a conclusion when omitted. A denominator or group label matters as much as the numerator.
Review decimal separators and units in context. Do not convert a number's notation simply because it looks unfamiliar in English. Compare with the table or method definition before deciding whether it is a punctuation error or a different convention.
For an illustrative review entry, write: “Table 2, intervention column: check the value and unit against the source; translated discussion appears to describe the control group.” Replace that example with the actual observed discrepancy and location.
Explain one figure in your own words
Select a figure central to the conclusion. Using the translation, identify what the axes show, what each series represents, which conditions were compared, and what the caption says. Then repeat with the original.
A difference reveals where to look: the plot, legend, caption, or prose. Preserved picture pixels cannot compensate for a mistranslated axis label or an omitted note about the sample.
For tables, read column headings and units before interpreting values. For equations, check signs, indices, numbering, and the sentence defining the variables. The formula and footnote checklist helps distinguish recognition problems from language problems.
Trace citations back to the actual works
Follow several citations from body text to the reference list, including one attached to a claim you plan to quote. Verify authors, year, title, and identifier against the source paper or the cited work.
Do not accept a reference merely because it is neatly formatted. A wrong year or normalized author name can make a different work appear plausible. Check that a note marker still refers to the right qualification rather than a nearby citation.
For an arXiv source, preserve the version you used. If another reader retrieves a different version, page and equation locations may differ even though the general title is the same.
Separate a reading copy from a submission-ready manuscript
PDF Craft can produce a translated PDF or other supported output for reading and further work. It does not certify the science, approve a citation, or replace a qualified editor.
If you intend to submit the text, consult the journal's current author instructions, citation style, and disclosure requirements. Have unresolved scientific meanings checked by a reader with the relevant subject and language knowledge. Do not represent unchecked output as professionally reviewed.
Finish the review by listing unresolved issues and corrected passages. Keep the original export as well as your edited copy. That makes the review reproducible: someone can see which claim was checked, against which source, and why the wording changed.