The preprint that grew up

You cited a preprint in your second year because it was the only version there was. It has since been peer-reviewed, revised and published in a journal. Your bibliography still points at the version that existed before any of that happened.

This is the most common reference-currency defect in doctoral work and one of the easiest to fix — provided anything tells you it happened, which is where most reference workflows fall down.

What changed, and what did not

An invented but ordinary case. A candidate cites Ferreira, N. (2022), a preprint on a public server, for a result she builds a chapter on. Eighteen months later the same work appears in a journal: peer-reviewed, revised in response to that review, paginated, with a publisher DOI of its own.

Both versions now exist, permanently. Neither one is wrong. They are, however, not the same document.

  THE PREPRINT                   THE PUBLISHED ARTICLE

  Ferreira, N. (2022)            Ferreira, N. (2023)
  preprint server                Journal of Applied
                                 Hydrology, 41(2), 118-134

  DOI resolves — permanently     DOI resolves — the version
                                 of record

  the version you read           the version peer review
                                 changed

  no volume, no pages            citable by page

  year: 2022                     year: 2023

Look at the last line. The year moved. That is the detail that quietly breaks things downstream, because an author-date citation is built from the year — every (Ferreira, 2022) in your text is now pointing at a version that is no longer the one you would want an examiner to read.

A preprint that is later published does not become wrong. It becomes the earlier of two versions, and you are citing the earlier one by accident rather than by choice.

Why nothing tells you

Every check passes, exactly as with a retraction. The preprint DOI resolves — preprint servers keep records permanently and correctly. The entry parses. The citation pairs with the entry. The metadata is complete and accurate.

The only signal that anything has changed lives in the relationship between two records in a bibliographic database, and nothing in an ordinary reference workflow ever looks at it. Your reference manager will not tell you: it stored what you gave it, and what you gave it is still true.

What the update pass actually does here

This is the case the update branch is built for, and the mechanism is not clever — it is just a search that nobody normally runs.

Each matched reference is queried against the bibliographic databases for a newer record describing the same work. A published version of a preprint is precisely that: same authors, same title or a close variant, later date, and in most cases an explicit relationship recorded by the publisher linking the two.

What comes back is a proposal, not a change. The published record is shown beside your existing entry with what differs, and you accept or reject it. Where no published version exists — which is often, since a great many preprints never are published — the tool says so and the preprint stays. That is a real answer, not a failure.

  reference 61                   a newer record exists

  your entry    Ferreira, N. (2022). Catchment
                response modelling. Preprint.

  proposed      Ferreira, N. (2023). Catchment
                response modelling. Journal of
                Applied Hydrology, 41(2), 118-134.

  because       same authors, same title, later
                date, and the publisher records
                this as the published version

  ACCEPT   REJECT           and 4 in-text citations
                            move from 2022 to 2023

The part people get wrong: the in-text citations

Accepting the new entry is the easy half. The half that causes trouble is that (Ferreira, 2022) appears four times in your text, and every one of them must become (Ferreira, 2023).

Done by hand, this is where the errors enter. Find-and-replace catches three of the four because the fourth was written (Ferreira 2022) without the comma, or sits inside a longer bracket alongside two other citations, or was typed in a table caption. The bibliography is then correct and the text disagrees with it — which is a worse state than before, because now the mismatch is invisible to you and obvious to a checker.

So an accepted replacement carries its occurrences with it. One decision, applied to every place the work is cited, with the count shown before you commit to it. The same principle as updating an ageing bibliography generally: the unit of change is the work, not the line.

How the title changes, and why that matters for matching

Peer review frequently rewrites a title. Not beyond recognition — but enough that a naive string comparison between your entry and the published record would not call them the same work.

  preprint    Catchment response modelling in
              semi-arid basins

  published   Modelling catchment response in
              semi-arid basins: a comparative
              approach

  shared      the authors, the subject, and the
              publisher's own record linking
              the two versions

This is why the search does not rely on the title alone. Authors, subject and the publisher-recorded relationship between the preprint and its published version all contribute, and where the relationship is explicitly recorded it is by far the strongest signal — it is the publisher stating that these are one work.

It is also why the proposal shows you what differs rather than simply asserting a match. A retitled paper is exactly the case where you want to read the two lines side by side before agreeing, because you are the one who knows which paper you actually read.

When to keep the preprint on purpose

Three cases, and all of them are legitimate.

If you are citing something the preprint said that the published version does not — a figure cut in review, an analysis revised — then the preprint is your source and swapping it would misrepresent what you read. Say so in the text.

If your work is about preprints, or about how a specific result changed through review, the preprint is the object of study.

And if the published version sits behind a paywall your readers cannot pass while the preprint is open, some supervisors will ask you to cite both. That is a disciplinary convention, not a referencing error, and it varies enough that your department is the authority rather than any tool.

What this does not claim

Citing a preprint is not a defect and this post is not an argument against preprints. In fast-moving fields they are the literature, and a thesis that ignored them until publication would be a year behind its own subject.

The defect is narrower: citing the earlier of two versions without knowing the later one exists. That is a currency problem, of the same family as the five-year rule myth — and like every finding in the report, it speaks to referencing quality and carries no suggestion of misconduct. It is what happens when a document takes four years to write and the literature does not wait.