Updating a reference without breaking your numbering

In an author-date thesis, replacing a reference changes one entry and a few brackets. In a numbered thesis, replacing a reference near the front can renumber the rest of the document.

Vancouver, IEEE and their relatives number references in order of first appearance. That makes the number a property of position, not of the reference — which is the fact that makes updating them dangerous, and it is not obvious until it has already gone wrong.

Why one change moves everything after it

Take a chapter citing forty works. You discover that reference 7 was a preprint that has since been published, and you want the published version.

If the replacement is simply swapped in place, nothing moves — the new work inherits position 7. But that is only true if the replacement is cited in the same place. The moment you add a source, or remove one, or the replacement first appears at a different point in the text, the ordering changes and every number downstream shifts by one.

  BEFORE                    AFTER REMOVING [7]

   [6] Adeyemi 2019          [6] Adeyemi 2019
   [7] Ferreira 2022  <-     [7] Okonkwo 2020
   [8] Okonkwo 2020          [8] Marchetti 2016
   [9] Marchetti 2016        [9] Ibarra 2020
  [10] Ibarra 2020          [10] Vidmar 2018
  [11] Vidmar 2018          ...

  every number from 7 onward is now
  attached to a different work — and
  every [n] in your text still says
  what it said before

The bibliography is correct. The text is now wrong in thirty places, and each error is a citation pointing confidently at the wrong source. Nothing is malformed, so nothing complains.

In a numbered style the reference list is not a list of your sources. It is a list of your sources in a particular order, and the order is load-bearing.

Why manual updating goes wrong so reliably

Because the work scales with the size of the document rather than the size of the change.

An author-date correction is local: change (Ferreira, 2022) to (Ferreira, 2023) wherever it appears, and every other citation in the thesis is untouched. A numbered correction that shifts the order is global: one decision, and potentially every subsequent bracket in three hundred pages.

People know this, which produces the second failure mode — they avoid updating at all. A reference they know is superseded stays because renumbering feels more dangerous than the outdated citation. That is a rational response to a real risk, and it is why numbered theses often carry older bibliographies than author-date ones.

What the tool does about it

Two things, and neither is clever — they are just the things you cannot do reliably by hand.

The numbering is rendered, not stored. Your references are held as structured records, and the numbers you see are produced from the style on demand. Change the set or the order and the numbering is regenerated from what is actually there. There is no manually maintained sequence to drift out of step, because there is no sequence written down.

An update carries its occurrences. When you accept a replacement, every place the work is cited moves with it — the count is shown before you commit, so you know whether you are changing one bracket or eleven. The unit of change is the work, not the line.

  reference [7]           a newer record exists

  your entry   Ferreira, N. (2022). Preprint.
  proposed     Ferreira, N. (2023). J. Appl.
               Hydrology, 41(2), 118-134.

  in the text  cited 4 times — [7] throughout

  numbering    unchanged: the replacement is
               cited first at the same point

  ACCEPT   REJECT

That last line is the one to read. When a change would renumber, it says so before you accept rather than after — which is the difference between a decision and a discovery.

Switching style late is the same problem, larger

Every so often a department asks for a different style after the thesis is written — Harvard to Vancouver, most painfully, because it converts every author-date bracket into a number and imposes an order that did not previously exist.

Done by hand this is days of work and produces errors at a rate proportional to fatigue. Done from structured records it is a re-render: the same references, formatted by a different set of rules, with the text itself untouched. The count of what will change is shown and confirmed first, because rebuilding every entry in a document is not something to do by accident. There is more on why styles diverge so much in the APA default that broke every Vancouver thesis.

The range that hides a broken citation

One numbered-style detail deserves its own warning, because it survives casual proofreading better than anything else: the collapsed range.

Numbered styles compress consecutive citations. Citing works 4, 5, 6 and 7 together does not produce [4,5,6,7] — it produces [4-7]. Which is tidier, and which means the bracket no longer names its members.

  you wrote        [4-7]
                   meaning Adeyemi, Okonkwo,
                   Marchetti and Ferreira

  after a change   [4-7]
                   now meaning Adeyemi, Okonkwo,
                   Marchetti and Ibarra

  the bracket is unchanged, correctly
  formatted, and cites a work you never
  intended to cite there

Proofreading cannot catch this. The bracket looks right because it is right as a piece of formatting — there is simply no way to tell from [4-7] which four works it now stands for. Only expanding the range against the current list answers it, which is exactly what a cross-check does and what reading the page does not.

Three habits for numbered theses

1. Never type a number. If a bracket in your text was typed by hand rather than generated, it will not move when the order changes. This is the single largest source of numbered-citation errors.
2. Do your updating in one pass. Renumbering once at the end is one event to check. Renumbering six times across a month is six opportunities for the text and the list to drift apart.
3. Re-run the cross-check after any renumber. It is the only thing that confirms every bracket still points at the entry you think it does — and in a numbered style that is precisely the property nothing else verifies. A mangled entry compounds this, as in the anatomy of a mangled author.

What this does not claim

Numbered styles are not worse than author-date ones. They are more compact, they read cleanly in technical prose, and in many fields they are simply what is required. The trade is that they move the fragility from the entry to the ordering, and the ordering is invisible.

And a numbering mismatch is not a finding of misconduct or a judgement about care. It is the predictable result of editing a document whose citation numbers depend on the position of everything before them — which is a property of the convention, not of the person writing in it.