Numbered citations that point at nothing

In an author–date style, a citation carries its own identity: (Okafor, 2021) says who and when. In a numeric style it carries none. [12] means "the twelfth thing in the list at the back", and every property you might want from a citation — who wrote it, when, whether it exists — lives somewhere else entirely.

That makes numeric citations compact, elegant to read, and vulnerable in ways author–date citations simply are not.

Four ways a number stops pointing at the right thing

1  DRIFT
   an entry was inserted or deleted and the list
   renumbered; the text did not
   → [12] now means a different paper
   → nothing looks wrong anywhere

2  DANGLING
   [47] in a list of 44 entries
   → the only failure that is obvious, and the
     rarest

3  A GAP IN THE LIST
   entry 23 was removed; the list runs
   …22, 24, 25…
   → every citation above 23 is now ambiguous:
     did the numbers shift or did they not?

4  NEVER CITED
   entry 31 exists and no [31] appears anywhere
   → an orphan, invisible in a numeric style
     because nobody reads a numbered list

Only the second announces itself. The first is the dangerous one: a drift produces a document in which every citation resolves to something, so nothing is broken and everything after the insertion point is wrong.

In an author–date style a broken citation looks broken. In a numeric style it looks like a citation to a different paper.

Square brackets and parentheses are not equivalent

This is the part that surprises people, and it is a genuine asymmetry rather than a stylistic preference.

[12] is unambiguous. Square brackets appear almost nowhere else in academic prose, so a bracketed number is a pointer and can be treated as one with confidence — it can be given a stable identity and matched one-to-one against the twelfth entry.

(12) is not. Parentheses around a number are used for a great many things, and a number in parentheses is indistinguishable, on its own, from several of them.

(12)   a citation to reference 12
(12)   a truncated year — "(2012)" with the
       century lost in extraction
(12)   a quantity, a count, a sample size
(12)   an equation or figure number

Compare:

[12]   a citation to reference 12
       ...and essentially nothing else

The bracket carries the meaning. The parenthesis
does not.

The practical consequence is measurable inside a checking pipeline: bracketed citations can be anchored to their list positions directly, while parenthesised ones have to be handled through a weaker, date-shaped route, because a bare number in parentheses cannot be distinguished from a year fragment without reading the sentence around it.

None of which is your fault if your style uses parentheses — several respectable ones do. It is worth knowing because it explains why a check may be more confident about a bracketed thesis than a parenthesised one, and why the latter benefits more from a human eye.

Ranges and groups

Two compact forms cause disproportionate trouble.

A range. [13–17] is five citations written as one, and every one of them has to exist. It is also the form most likely to survive a renumbering unnoticed, because the visible text changes by one character while the meaning changes by five references.

A group. [3,9,12] is three independent pointers sharing a bracket. If entry 9 is deleted, the group still reads perfectly and now names a different middle source.

Both are worth expanding mentally when you check. A range of five is five things to verify, and treating it as one is how a thesis ends up with a citation to a paper that was removed from the list two years ago.

How to check the numbering yourself

You do not need to verify every citation. Numbering drift has a signature, and three checks find almost all of it in twenty minutes.

One: check the highest number. Find the largest citation number in your text and count your reference list. If the list is shorter, you have a dangling pointer. If it is longer, the surplus sits at the end of the list — which is exactly where orphans collect.

Two: check for gaps in the list. Read down the numbers in your bibliography. Any break in the sequence means an entry was removed after numbering, and every citation above the gap is suspect.

Three: spot-check by content, at three depths. Take a citation from early, middle and late in the document, look up the entry it names, and ask whether that source plausibly supports that sentence. Drift is cumulative — it starts at an insertion point and affects everything after — so an early citation that resolves correctly and a late one that does not tells you roughly where the problem began.

Chapter 2   [7]  → entry 7 is on-topic       ✓
Chapter 4   [23] → entry 23 is on-topic      ✓
Chapter 6   [41] → entry 41 is about
                   something unrelated       ✗

The break is between chapter 4 and chapter 6.
Something was inserted or removed there, and
every citation after it is shifted.

Three lookups located it. Verifying all 210
would have found the same thing.

What to do when the numbers have drifted

Do not renumber by hand. The temptation is to fix the visible symptom — change [41] to [38] — but if the drift affects everything after an insertion point, hand-fixing produces a document where some citations were corrected and others were not, which is worse than a consistent shift.

Rebuild the numbering from the text. In a numeric style the correct order is first appearance in the document, so the list can be regenerated from the citations rather than the other way round. If your citations are live fields, a reference manager does this. If they are typed text, it is the same laborious operation as a style conversion, and worth the same care.

And check before converting away from numbers. If you are moving to an author–date style, verify the numbering first — because after conversion the numbers are gone and a drift becomes a set of confident, named, wrong attributions.

That last consequence deserves plain statement. A drifted citation does not merely point at the wrong row; it attributes a finding to researchers who did not produce it. An undetected drift is bad luck and nobody's fault — it is a mechanical error with a mechanical fix. Submitting after you have seen it is a different matter: at that point the attribution is one you know to be false, and knowingly misattributing a finding to a researcher who did not make it is academic misconduct rather than a formatting slip.

Why numbering drifts in the first place

Worth understanding, because the causes are avoidable and the symptom is not.

Typed citations with a managed list. The commonest cause by a distance. Your reference manager regenerates and renumbers the bibliography whenever entries change; your in-text [12] is plain text and does not. Every insertion silently invalidates everything after it. If your citations are live fields this cannot happen — and if they are typed, it happens every time.

A late deletion. Removing one source you decided against is a one-second edit with a document-wide consequence in a numeric style, and it is the edit people are least likely to remember making.

Merging chapters or co-authored sections. Two documents with independent numbering, combined. Both sets of numbers were correct; neither is now.

And a converted file. Fields become plain text when a document moves between formats or is opened in a different editor — and the citations that were live and self-maintaining silently become characters that are not.

AUTHOR–DATE
  delete an entry
  → the citations to it become unmatched
  → a check finds them: "no entry for this"
  → the damage is LOCAL and visible

NUMERIC
  delete an entry
  → the list renumbers
  → every citation after it still resolves
  → the damage is GLOBAL and invisible

Same edit. One produces a findable error, the
other produces a document that reads perfectly
and means something else.

Which is the argument for checking numeric documents more often rather than less. In an author–date thesis a defect waits patiently to be found. In a numeric one it propagates, and the longer it sits the further from the original edit you are when you notice.

The mirror-image problem

Everything above is about a citation that reaches the wrong entry, or none. The opposite failure exists too: text that looks like a citation and is not — a bare (12) that was a sample size, or a year in prose that a checker decides to treat as a reference.

They are worth thinking about together, because a numeric style suffers from both at once. The bracket form is safe from the second and exposed to the first. The parenthesis form is exposed to both — which is the strongest practical argument for square brackets in any thesis that has a choice.

If your handbook permits either, take the brackets. It costs nothing, reads identically, and removes an entire class of ambiguity from a document you will spend four years editing.