Research & PolicyFree
Preprint vs Published Version Comparison
Send the preprint you cited and the published paper, get a change register and the effect on every sentence resting on it.
River reads both versions and builds a change register: one row per discrete difference, typed by whether it is wording, a changed number, an altered analysis, an added result or a removed one. Then it does the part a diff tool cannot. For every sentence in your own draft that rests on the work, it says whether the sentence survives unchanged, survives with a different number, or loses its support entirely. The register and the sentence-level verdicts arrive together, with the passage from each version quoted side by side.
Search this and you get generic advice about preferring the published version. None of it names the trap. bioRxiv's own policy page states that revised versions of manuscripts retain the same DOI assigned to the first version, and that citing a specific version means adding the version-specific URL. A bare DOI therefore resolves to whatever is current, not to what you read. The citation still works, still returns a real paper, and now says something else.
Written for the author revising after review, the reviewer checking whether a cited preprint holds, and the doctoral student whose literature review is two years old. It sits beside a claim-level check of every citation in the draft, which finds the problem this run resolves. Where the version question is one input to a bigger read, an evidence map with a stated verdict and a synthesis organised around disagreement are where it lands.
The finding survived and the number did not
A preprint posted in 2024, cited in a draft, and published fourteen months later in a journal. The change register holds 34 discrete differences across the two versions. Fourteen are wording that touches no claim. The other twenty are substantive: eight reported numbers changed, five analysis choices changed, four results appeared that the preprint did not contain, and three were removed. Fifty-nine per cent of the differences matter, which is not what a reader expects from a version bump.
The headline effect moved from 0.41 to 0.28, a fall of 31.7 per cent, because the analysed sample dropped from 1,842 to 1,504 after the journal required two exclusion criteria the preprint did not apply. That is 338 participants, 18.3 per cent of the original. The confidence interval moved from a span of 0.19 to 0.63 into a span of 0.04 to 0.52. It widened rather than narrowed, which is what a smaller sample does to an interval.
Both intervals exclude zero, so the finding did not reverse and no referee can say it did. Each version's point estimate also sits inside the other's interval, so the two results are statistically compatible. What changed is the number, and the draft quoted the number. Of six sentences resting on this work, two survive untouched, three need 0.41 replaced by 0.28, and one rested on a result the published paper removed. Four of six need an edit, and none of them looks wrong on the page.
How it works
Send both versions
The preprint you cited and the published paper, plus the draft that rests on them.
Build the register
One row per difference, typed by whether it is wording, a number, or an analysis.
Propagate the changes
Each sentence in your draft checked against the version the citation now leads a reader to.
Return the verdicts
Which sentences survive, which need a new number, and which lost their support altogether.
What you get
- One row per difference between the versions, typed by whether it touches a claim
- Every sentence in your draft that rests on the work, with a verdict on each
- The version your citation actually resolves to now, which is rarely the one you read
- Changed numbers isolated from changed wording, because only one of them reaches your sentences
- Results the published version removed, named, since a citation to them cannot be repaired
- Whether the finding reversed or only the magnitude moved, stated as two separate answers
Common questions
Why not just cite the published version and move on?
Because your sentences were written against the preprint, and swapping the reference does not update them. A reference swap makes the citation resolve to a paper that no longer contains the number quoted above it, which is worse than the original problem because it now looks checked. The sentences are the work, not the reference list.
How do I know which preprint version I actually read?
Check your own notes first, then your download date against the posting dates of each version. Where neither is available the run says the version is unestablished and compares against the earliest one, which is the conservative choice. It also records that the ambiguity exists, because a reviewer can ask and you should have an answer.
Should I record the version in my citation going forward?
Yes, and at least one major funder requires it. The NIH notice on interim research products tells applicants to list any information about the document version alongside the identifier and the object type. A version-specific URL does the same job in a reference list, and it makes this whole reconciliation unnecessary next time.
What if the two versions agree?
Then the run says so in one line and you stop. That outcome is common and it is worth the ten minutes, because the alternative is a referee finding the one case where they disagree. The register still records what changed in wording, which is occasionally useful when a reviewer asks why you quoted an older phrasing.
Can it tell me whether the finding reversed?
It answers that separately from whether the number changed, because the two are different questions and conflating them is how a version note becomes overstated. A point estimate can fall by a third while both intervals still exclude zero. The run states the direction verdict and the magnitude verdict as two lines, and never lets one stand in for the other.
Does it work when the preprint was never published?
Partly. It compares the posted versions against each other, which catches revisions made after you cited it, and reports that no journal version exists. An unpublished preprint two years past posting is itself worth a note in the draft. Where the work matters to a conclusion, a documented search for a later version is the follow-up.
What do I get back?
A Sheet holding the change register, one row per difference with its section, its type and the text from both versions. A Doc stating the substantive differences in method, sample and findings, then the verdict on each sentence in your draft, with the replacement number where one exists.
Preprint vs Published Version Comparison
Fill in the form and your workspace opens with the work already underway.