The DOI that pointed at the wrong paper

A DOI is the strongest identifier in your bibliography and the one nobody ever checks. It is a working link, so it looks correct — right up to the moment somebody clicks it and arrives at a different paper.

This defect is rarer than a mangled author name and considerably more embarrassing, because the reference looks impeccable. Everything about the entry is right except the one field that claims to identify it beyond doubt.

How a wrong DOI gets into a bibliography

Almost never by carelessness with the DOI itself. It arrives attached to something else.

  THE ADJACENT ROW
  copied from a results list, one line
  off — the DOI of the paper above

  THE ISSUE, NOT THE ARTICLE
  some publishers mint a DOI for the whole
  issue; it resolves, to a table of contents

  THE MANAGER MERGE
  two records merged in a reference manager,
  one record's title kept, the other's DOI

  THE CORRECTED VERSION
  the DOI of a correction notice rather than
  of the paper the correction concerns

Every one of those produces a syntactically valid DOI that resolves successfully. There is nothing malformed to detect. The string is real; it simply belongs to something else.

Why it survives every check that matters

Here is the uncomfortable part. A DOI is treated as an identifier — the strongest kind of evidence a matcher has — so a wrong one does not merely go unnoticed. It is actively trusted.

In our own matching cascade, a shared identifier is the first pass: if the text and the entry name the same DOI, the pair is settled and no later stage revisits it. That ordering is deliberate and we would not change it, because an identifier really is the best evidence available. It also means a wrong identifier is inherited quietly by everything downstream. We wrote about why the cascade runs in that order in the anatomy of a mangled author.

A wrong identifier is not a weak signal. It is a strong signal pointing in the wrong direction, which is why it needs a different kind of check.

The check that catches it

Not a matching rule — a resolution. The verification pass follows each DOI out to the publisher's own page and compares what the DOI lands on against what your entry says.

That comparison is the whole mechanism, and it is only possible because the link is actually followed rather than merely validated as well-formed.

  YOUR ENTRY

    Ibarra, L. (2020). Sediment transport
    under variable flow. Water Resources
    Review, 33(4), 210-229.
    doi:10.xxxx/wrr.2020.0447

  WHERE THAT DOI LANDS

    Okafor, T. & Ibarra, L. (2020).
    Bedload sampling methods.
    Water Resources Review, 33(4), 188-209.

  the journal matches · the issue matches
  the year matches · one author matches
  THE PAPER DOES NOT

Note how much agrees. Same journal, same issue, same year, an overlapping author — which is exactly what you would expect from the adjacent-row error, and exactly why a check that compared only some fields would call this a match. The title and the pages are what disagree, and they are the fields that identify the actual work.

What the report says, and what it does not do

It reports the disagreement and stops there. The entry is flagged as resolving to a different paper, both descriptions are shown, and the link is there for you to open.

It does not silently correct the DOI, and it does not silently correct the entry — because at that point there are two possibilities and software cannot distinguish them. Either the DOI is wrong and your entry describes the paper you meant, or the DOI is right and your entry's title and authors were copied from the wrong place. Those need opposite fixes, and only you know which paper you actually read.

This is the same discipline as everywhere else in the product: it establishes a fact, presents it, and leaves the judgement with the author.

The variant that is not a wrong DOI at all

One case looks identical in a report and needs a completely different response: the DOI is correct and the destination has moved.

Journals change hands. A title is sold, a society moves its archive to a different publisher, a platform is retired. The DOI system is built for exactly this — the identifier is permanent and the address it points to is updated — but the update is not always immediate, and during the gap a resolution can land on a holding page, a search result, or a login wall that names nothing.

  A DIFFERENT PAPER
  the identifier belongs to something
  else — a real defect, fix the entry

  A HOLDING PAGE
  the archive moved; the record exists
  but the address is catching up

  A LOGIN WALL
  the paper is there and the page will
  not say so without a subscription

Only the first is a defect in your bibliography. The second usually resolves itself within weeks and is worth re-checking rather than editing. The third is not a problem at all — your reader at a university with a subscription will see the paper your entry names.

This is why the report shows you the destination rather than issuing a verdict on it. Three outcomes that look the same to an automated check are obvious to a person who opens the link, and telling them apart takes seconds.

Why examiners notice this one

Most referencing defects are invisible to a reader. A missing entry requires cross-checking the whole bibliography; an inconsistent style requires knowing the style. A wrong DOI requires one click, and it is the click a curious examiner makes when the cited claim is central and they want to read the source.

They land on a different paper. And the natural reading — however unfair — is not "a copy-paste error three years ago". It is "did they actually read this?"

That gap between the trivial cause and the severe impression is why this defect class is worth a specific check rather than trusting to luck. Fabricated references produce the same impression by a different route, which is covered in what to do when the citations don't exist.

What this does not claim

A wrong DOI is not evidence of anything about the candidate. The overwhelming majority arrive through the four routes above, all of which are mechanical, all of which happen while doing the work correctly, and none of which involves misconduct of any kind. The most common single cause is a results list read one row out at eleven at night.

Nor does resolution prove a reference is right. A DOI landing on the paper your entry names tells you the identifier is correct; it says nothing about whether the paper supports the claim you cited it for. That judgement was always yours, and no reference checker is in a position to make it.