The bare-year bug

"As Farah and Lindgren (2019) showed, burnout tracks staffing gaps — an effect that strengthened in their follow-up study (2020)." Perfectly good academic English. And sitting at the end of it, one of the most quietly troublesome objects a citation checker ever meets: a citation with no name. Whose is "(2020)"?

A human reads it without noticing there was a question. The year obviously belongs to Farah and Lindgren; the sentence says so. But "the sentence says so" is doing real work — work that software has to do explicitly, and that several plausible-looking shortcuts do wrong. This case file dissects the bare year: the three ways a tool can read it, the rule that reads it correctly, and a bug of our own in this exact class — because a case file about reading citations honestly should be honest about its author too.

Three readings, two of them wrong

…Farah and Lindgren (2019) showed …
   strengthened in their follow-up (2020).

reading A — a citation of nobody
  the checker hunts the list for "(2020)"      phantom missing ref

reading B — decoration, skip it
  "(2020)" is treated as noise                 a real source unchecked

reading C — inherit the anchor
  authors carried from Farah & Lindgren (2019)
  year taken from the token                    Farah & Lindgren, 2020

names invented; the construction is textbook author–date style

Reading A — treat every parenthesised year as a citation — floods the report with phantoms. The reference list has no entry called "(2020)", so the checker declares a missing reference that never existed, and it does so for every bare year in the document. Reading B — ignore years without names — fails in the opposite, quieter direction: the 2020 paper is genuinely cited, deserves an entry in the list, and deserves checking; skipping it means a real reference can rot or go missing without any row ever saying so. Silent under-counting is the worse failure, for reasons this blog keeps meeting: what you can't see, you can't fix.

Reading C is the correct one, and its correctness has conditions. The bare year must be directly continuing the anchor citation's discussion — same sentence or an unbroken clause, no other author intervening. Then, and only then, the year inherits the anchor's authors and becomes a full citation in its own right: Farah & Lindgren, 2020 — a second paper, expected in the reference list on its own line.

A bare year is a dependent record: it borrows its meaning from an anchor. Handle the anchor carelessly and the dependant breaks in silence.

The leash on the rule

Inheritance needs a leash, because the pattern "name … year" appears in prose that cites nothing. "Yuchtman and Seashore in 1967 defined effectiveness as…" is history-telling, not citation — and a rule greedy enough to capture it would turn every date in every literature review into a phantom reference. So the inheritance rule runs narrow by design: a parenthesised year, directly continuing an explicit anchor citation, with nothing in between that could re-anchor it. Doubt does not widen the net; doubt routes the token to the chaff drawer, where a human can see it and decide. We never trade a visible question for an invisible guess.

The bare year's family

Once you know to look, the trailing year turns out to be one member of a small family of constructions that all pose the same question — which authors does this year borrow? — with escalating difficulty:

• The semicolon run: (Farah & Lindgren, 2017; 2020). Two years inside one parenthesis, one author list. Standard style, two works, two expected entries — the inheritance is explicit and safe.
• The suffixed pair: (Farah & Lindgren, 2020a, 2020b). Same year twice, distinguished by suffix. Here the reference list carries the burden: two entries whose only difference may be the suffix letter, which makes suffix-preserving matching non-negotiable — fold the letters and two works become one.
• The prose continuation: "…as their follow-up study the next year confirmed." No parenthesis at all. A human reads a citation; a tool must not, because there is no token to anchor — this is where the citation ends and prose begins, and the honest treatment is a flag for the writer, not an invented record.
• The interrupted chain: "(Farah & Lindgren, 2019) … (Okafor, 2021) … a view revisited in (2020)." An author intervened. The bare year's anchor is now genuinely ambiguous — the nearest citation is Okafor, the plausible owner is Farah & Lindgren — and any rule that picks silently will be wrong often enough to matter. This one goes to a human, with both candidates shown.

The family matters because each member fails differently under a greedy rule and differently again under a timid one. A checker has to hold the whole set — inherit where the text licenses it, refuse where it doesn't, and put the boundary cases in front of you rather than under the rug.

Our own bug in this class

Here is the part the textbook version skips: the dependency in "dependent record" cuts both ways, and we learned it the usual way. In an earlier version of our sister product, a bare year correctly inherited its anchor's authors — once. But records get edited: suppose the anchor citation is later corrected to a different author entirely, say (Okafor, 2019), because the writer had the wrong lead name. The trailing "(2020)" kept riding with whatever it had inherited first. The sentence now read Okafor; the hidden record still said Farah & Lindgren; and one rendering path — an annotated view that had its own way of drawing the text — had never run the splitting rule at all, so it showed the stale pair to the user.

The fix encoded two lessons we now treat as law. First, a dependent record must re-qualify whenever its anchor changes: if the lead citation's author is updated, the trailing year stops borrowing and is re-written with its own author, visibly. Second, every door runs the same rules: any view that renders citations — the working screens, the annotated document, the export — must pass through the same splitting logic, because a rule enforced in three places out of four isn't a rule, it's a habit. The repaired behaviour shipped to both products, and the class joined our regression corpus so it cannot quietly return — the same discipline that governs every count we show.

What the examiner sees

It's worth being concrete about the stakes, because the bare year feels like a punctuation nicety. Suppose the 2020 entry never made it into the list — the most common outcome, since the writer's attention was on the anchor. An examiner reading chapter 2 meets a claim resting on "their follow-up (2020)", goes to the list, and finds only 2019. Now they are holding exactly the question the construction buried: is there a second study at all? A claim that had two papers behind it suddenly reads as a claim with one paper and one assertion — and the repair, at viva or review stage, costs a correction round instead of the six words it would have cost in draft. Every ambiguous construction in a document transfers work from the writer, who could resolve it instantly, to a reader who can't. The whole point of checking before submission is to make that transfer never happen.

What this means for your own draft

1. The bare year is legal — make sure both papers are in the list. The construction cites two works. Writers add the 2019 entry and, often, forget the 2020 one; that's a real missing reference an examiner will chase.
2. Re-qualify in the final pass. "…their follow-up study (Farah & Lindgren, 2020)" costs six words and removes the ambiguity for every reader and every tool that meets your document after you.
3. Watch what editing does to the pair. If you correct the lead citation's authors, the trailing year's meaning changes with it. Re-read the whole sentence after any author fix — the failure mode we shipped is one you can commit by hand just as easily.
4. Adjacent years, same authors? Check whether the two are truly two works. A 2019 preprint and its 2020 published version are one work that grew up — cite the version of record.