Link rot and DOI rot

You cited a paper in your second year. You submit in your fifth. In between, you did not touch that reference — and it stopped working anyway.

This is the failure mode nobody plans for, because every other defect in a bibliography is something you did. Link rot is something that happens to a document that has been sitting still.

What actually rots, and how fast

Not everything decays at the same rate, and knowing the order tells you where to spend a re-checking hour.

MOST FRAGILE
  a bare URL to an organisation's page
    → site reorganisations, rebrands, CMS migrations
    → the commonest single cause of a dead reference

  a URL to a news article or blog post
    → paywalls appear, archives are pruned

  a preprint server link
    → often superseded rather than dead: the work
      moved to a journal and the preprint became
      the wrong thing to cite

  a DOI
    → designed to survive all of the above
    → and it still fails, in four specific ways

MOST DURABLE
  a book with an ISBN, cited by edition
    → nothing to resolve, nothing to break

The pattern is that fragility tracks how much of the citation depends on somebody else's infrastructure staying the same. A book citation needs nothing from anyone. A URL needs an organisation to keep a particular path working across four years of website redesigns, which is not something anyone has promised you.

DOIs rot too, in four ways

A DOI is supposed to be permanent, and mostly is. The failures are worth knowing because they look different from a dead link and are easy to misread.

1  THE PUBLISHER MOVED
   a journal is sold or transferred; the DOI record
   is updated but resolution is briefly inconsistent
   → often fixes itself. Re-check rather than act.

2  THE RECORD CHANGED UNDER IT
   the DOI resolves, but the metadata was corrected —
   a title typo fixed, an author added, a year amended
   → your reference is now subtly wrong, and nothing
     announced it

3  IT RESOLVES TO A NOTICE
   the work was withdrawn or retracted; the DOI now
   leads to the retraction rather than the paper
   → this is the identifier working correctly, and
     it is telling you something important

4  IT WAS NEVER RIGHT
   the DOI was mistyped or copied from a neighbouring
   record years ago
   → not rot. It never worked. You are only finding
     out now.

The fourth is the most common in practice and the least like rot, which is why it belongs here: a re-check is the only occasion on which anyone discovers it. A wrong DOI recorded in 2022 sits perfectly quietly until someone follows it.

The third is the one with consequences beyond your own bibliography. A retracted paper that is still cited as supporting evidence — with no acknowledgement that it was withdrawn — changes what your claim rests on. Retraction notices are frequently issued quietly, years after publication, and there is no mechanism that tells the people who already cited the work. Continuing to present a retracted finding as current evidence, once you know, is a straightforward integrity problem rather than a formatting one, and institutions treat it as academic misconduct.

Nothing in your document changed. The literature underneath it did, and it did so without telling you.

Why a doctorate is the worst case

Most writing has a short window between citing and publishing. A doctorate has the longest one in academic life, and the arithmetic is unkind.

Year 1   literature review assembled       ~180 refs
Year 2   fieldwork, few additions
Year 3   analysis chapters                  +140 refs
Year 4   discussion, revisions              +90 refs
Year 5   submission                          410 refs

The year-1 references are four years old at
submission. They were checked once, when they
were new, by someone who has since read
several hundred other papers.

Nobody re-reads them. That is the gap.

The compounding factor is that early references are usually the foundational ones — the theory chapter, the definitional sources, the works your framing depends on. They are simultaneously the oldest, the least recently checked, and the most load-bearing. A dead link in your final chapter is an inconvenience; a broken reference under your theoretical framework is a question in a viva.

When to re-check

Three moments, and the middle one is the one people skip.

When the literature review settles. Not because much will have decayed yet, but because this is when the list is small enough to fix cheaply and you still remember what each source was for.

Once a year, mechanically. Twenty minutes with a checker, or an afternoon by hand on the entries that matter. This is the step that turns a submission-week crisis into four unremarkable maintenance passes. Everyone agrees it is sensible; almost nobody does it.

In the month before submission. Not the week — the month. Finding that eleven references need attention is manageable with four weeks and miserable with four days, and some fixes require reaching a library or waiting on an interlibrary loan.

The fuller version of that final pass is a guide of its own, because ageing does more to a reference list than break links: preprints become articles, editions are superseded, and a source that was current when you cited it may simply have been overtaken.

What a failed resolution proves, and what it does not

Worth being careful here, because the temptation is to treat any failure as a verdict.

A DOI that does not resolve today might mean the identifier is wrong. It might equally mean the resolver was briefly unavailable, the publisher is mid-transfer, or the record is temporarily inconsistent. A single failed lookup is evidence, not proof — which is why an honest check reports "did not resolve" rather than "does not exist", and why we report an unresolved identifier separately from a missing entry rather than pooling them into one alarming number.

The practical response is proportionate. Re-check a failure a day later before doing anything. If it persists, search for the work by title — most "dead" references are alive at a new address. Only if the title search finds nothing at all does the question become whether the work exists, and that is a different conversation entirely.

The retraction problem deserves its own paragraph

Of everything in this post, retraction is the one worth a separate habit, because it is the only decay event that changes what your argument stands on rather than merely how a reader reaches it.

The asymmetry is stark. When a paper is retracted, the journal issues a notice and updates its record. Nobody contacts the people who already cited it — there is no mechanism that could, since citing is not a subscription. So a retracted finding continues to circulate through everyone who cited it before the notice and never looked again, which is the ordinary case rather than the exceptional one.

You cited it            2023, in good faith
Notice issued           2025, quietly
You submit              2026

Your text still says    "…as demonstrated by Alvarez
                         et al. (2021)…"

The DOI resolves        to a retraction notice
Your claim now rests    on a withdrawn result
Nobody told you         because nobody could

The fix is not dramatic. Sometimes the finding survives elsewhere and you re-cite a different source. Sometimes it does not, and the sentence changes. Occasionally the retraction is itself the interesting fact — if you are writing about the literature rather than from it, a withdrawn paper may be exactly what you want to discuss, cited explicitly as retracted.

What matters is that the decision is yours and taken knowingly. A retracted citation you did not know about is bad luck; the same citation left in place after you have seen the notice is a different thing entirely, and it is why an identifier re-check is the one maintenance pass genuinely worth doing on a schedule.

How to make a list that ages better

Prefer identifiers to addresses. Where a DOI exists, record it, even if you found the paper at a URL. The DOI is the thing designed to survive the next decade.

Record enough to re-find the work without the link. Full title, full author list, year, venue. A reference that can only be located through its URL is one broken address away from being unrecoverable — and a complete citation is searchable even when the address is gone.

Archive anything without an identifier at the time you cite it. Web sources, reports, standards, datasets. The snapshot costs a minute now and cannot be created retrospectively.

And note the date you last verified each entry, if only in your own notes. Not for the reader — for you, in year four, deciding which two hundred of your four hundred references you can safely skip.