When a university becomes an author

The in-text citation names two people: (Navarro & Iqbal, 2021). The reference-list entry names a building: School of Management, Bandari University. (2021). Employee advocacy and organisational voice… Same title, same year, same journal — but as far as any matcher is concerned, this citation is missing its reference and this entry belongs to nobody. The authors didn't vanish. They were evicted from their own byline by their employer.

Affiliation-as-author is one of the strangest-looking defect classes we meet, and one of the most mechanical once you know where it comes from. It is also a trap for checker design, because the obvious fix — "author fields should contain people" — breaks legitimate references on contact. Plenty of real authors are institutions.

How a department gets the author's chair

No writer types "School of Management" into an author field on purpose. The swap happens upstream, in the machinery that assembles references for us:

• PDF metadata. Publishers embed document properties, and some fill the author property with the corresponding author's affiliation, or with the typesetter's organisation. A reference manager importing "from PDF" trusts what it finds.
• "Cite this" widgets and scraped pages. Repository landing pages often list affiliation directly under the title. A scraper that reads the page visually — rather than from structured metadata — grabs the line in the author's position.
• Institutional reports done half-right. A working paper genuinely published by an institute gets a person-authored journal version later; the writer updates the title and year but the old corporate byline survives.
• Manual entry under deadline. The one everyone recognises: copy, paste, the affiliation came along, and no spell-checker will ever object.

The trap: institutions really do write things

The naive repair — flag any author that doesn't look like a person — detonates immediately, because corporate authorship is a first-class, style-sanctioned category. The World Health Organization authors guidance. The OECD authors reports. Standards committees author standards. A thesis citing (WHO, 2021) against an entry beginning World Health Organization is correct, and a checker must connect that pair, not complain about it — which is why our matcher carries a deliberate, disclosed bridge between an organisation's acronym in the text and its full name in the list. That bridge is mechanical and shown to you; silence would read as a wrong match.

The question is never "is the author a person?" It is "is this the name the work was published under?" — and that question has an external answer.

Telling eviction from authorship

What separates a leaked affiliation from a real corporate author is evidence, not vocabulary. The signals stack:

text cites          (Navarro & Iqbal, 2021)          people
list entry          School of Management, Bandari
                    University. (2021). Employee
                    advocacy and organisational voice…  a department

registry record     Navarro, D., & Iqbal, S. (2021).
(same title/year)   Employee advocacy and organisational
                    voice…                              people — named in the text

verdict             affiliation leaked into the author slot
proposed repair     restore Navarro & Iqbal            applies on your click

names and institutions invented; the class is common in imported bibliographies

The department-shaped string ("School of…", "Department of…", "Institute for…") raises the question; it never answers it. The answer comes from looking the work up in the registries — Crossref and its peers — by title and year. When the registry's record for that exact work lists people, and those people are the ones your text already cites, the case is closed by evidence three sources deep: your citation, the registry, and the entry's own title agree; only the author slot dissents.

When the evidence is that strong, the repair is deterministic and we propose it outright. When it's weaker — no registry record, a partial title match, a work that might genuinely be institutional — the case goes to the AI layer, whose job is judgement on candidates that already exist, and whose suggestions arrive marked as suggestions. Either way, nothing rewrites your entry until you click. A repaired byline is still an edit to your bibliography, and edits are yours to approve.

Corporate authors done properly, while you're there

Since the diagnosis forces you to think about organisational bylines, it's worth getting the legitimate ones right in the same pass — they have rules, and the rules are where marks quietly leak:

• Name the organisation as published, not as abbreviated. The list entry is World Health Organization, not WHO; the acronym belongs in the text, introduced at first mention in the style your guide prescribes — commonly (World Health Organization [WHO], 2021) first, (WHO, 2021) after.
• Don't let the same body appear under two names. An entry for World Health Organization and another for WHO is the organisational version of one author, two spellings — and it produces the same mirrored phantom findings.
• Sub-units cite under the parent unless the work says otherwise. A report issued by a ministry's statistics directorate is usually cited under the ministry, with the directorate in the entry where the style permits. If the title page names an author, the author outranks the issuing body.
• Alphabetise by the full name. Corporate entries sort as written — an entry filed under "WHO" sits in the wrong place the moment it's corrected to its full name, and Vancouver renumbering makes that a cascading edit rather than a cosmetic one.

Why this class earns deterministic handling

A fair question about the repair: if an AI layer exists, why bother with the deterministic path at all? Because this class, unusually, offers certainty — and certainty should never be spent through a probabilistic instrument. When the registry's record for the same title and year names the same people the text already cites, there is nothing left for judgement to do; every element of the conclusion is checkable, and the proposal can say exactly why it's right. The AI layer exists for the remainder — the working paper with no registry record, the title that matches at 80%, the genuinely institutional report. Keeping the two paths separate is what lets the evidence panel be honest: a deterministic repair shows you facts, an AI suggestion shows you reasoning, and you always know which kind you're approving. Collapsing them into one "smart fix" button would feel smoother and be strictly less trustworthy — the same trade we refuse everywhere the two kinds of confidence meet.

What to look for in your own list

1. Scan author slots for organisational grammar. "School of", "Department of", "Faculty of", "Centre for" at the start of an author field is worth thirty seconds each. Real corporate authors are usually the whole organisation, not a sub-department with a comma trailing into a university name.
2. Check the pair, not the entry. If your text cites people and the list answers with an institution — or vice versa — one of them is wrong, and the registry record for the title settles which.
3. Keep legitimate corporate authors consistent. Cite the same form the style prescribes (usually full name first mention, acronym after), and make sure the list entry uses the published byline, not your shorthand.
4. Fix the manager, not just the entry. If one import produced an evicted byline, its siblings from the same source probably did too — the same lesson as every imported-name defect.

And a word on how these arrive in numbers, because they rarely arrive alone. Affiliation-as-author is an import-batch defect: whatever produced one evicted byline — the PDF importer, the scraped landing page, the "cite this" widget — was probably used for a whole reading session, and its siblings entered the list the same evening wearing the same damage. When you find one, check the entries added around it. The efficient repair is by source, not by row: identify the import path, audit its cohort, and you'll usually clear three or four entries in the time the first one took to diagnose. This is also why the evidence panel shows you the pattern rather than just the fix — a writer who recognises the defect's fingerprint stops importing it.

One structural note: this whole class is invisible to identifier-based matching — a DOI doesn't care what the author slot says. That is a genuine argument for identifiers, with the caveat that registries are a source of truth about metadata, found through engines that each see a different slice of the record — which is exactly why we cross-check rather than trust any single field from any single source.