References¶
The encyclopedia cites 34,800 distinct works across 50,020 citations. 11,805 of those works (34%) have an identifier that resolves to a registered record. What that does and does not establish — and what is still unmeasured — is set out in how references were verified.
A further 314 entries have been withdrawn: 87 works the audit could not find in any authority, 225 inline prose definitions our extractor had stored as though they were sources and 2 duplicate records. Their pages remain and say why; they are excluded from every count below.
- Browse all 34,800 works — filter and sort by author, year, type, or how often each is cited
- The canon — the 250 most-cited works, effectively this project's reading list
- How references were verified — what the checking did and did not establish
Coverage, stated plainly¶
Reference density is very uneven across the four catalogs. That is worth saying outright rather than hiding behind a single total:
| Catalog | Artifacts | With references | Coverage | Citations |
|---|---|---|---|---|
| Primes | 1,525 | 1,420 | 93% | 15,562 |
| Domain-specific | 11,505 | 7,925 | 69% | 26,688 |
| Solution archetypes not yet referenced | 1,106 | 0 | 0% | 0 |
| Mechanisms | 9,858 | 7,723 | 78% | 7,770 |
Solution archetypes carry essentially no references today. That is a real gap, not a rounding error, and it is stated here rather than averaged away into a single headline number.
Checked against the source¶
Whether a work exists is one question; whether it says what the sentence attributes to it is the other, and the one that matters. Checking it means reading the source against the sentence, citation by citation, so it is slow work that will take a long stretch of calendar time to carry across the whole encyclopedia. So far 19,050 of 50,020 citations (38.1%), in the prime abstractions, domain-specific abstractions and mechanisms, carry a published grade from an automated reader that read them against the work's text or abstract; the grade appears under each citation on the article page and on the work's page.
The grades say how much of the sentence the source covers, not whether the sentence is true. Supported means the work states the claim. Supported in part means it states some of what the sentence asserts — most sentences make more than one assertion, and a single source rarely carries all of them; the rest is backed elsewhere or not yet checked. It is not a finding against the sentence. Borderline support means two runs of the same check disagreed. Not yet verified means exactly that: the default for most of the corpus today, and a gap in the record rather than a verdict. Could not be verified means a check was made but the source could not be read closely enough to judge the sentence: no full text or abstract could be obtained, or what could be read does not bear on it.
| Grade | Citations | Share | % |
|---|---|---|---|
| Supported1,228 works | 4,340 | 8.7% | |
| Supported in part4,196 works | 14,679 | 29.3% | |
| Borderline support21 works | 31 | 0.1% | |
| Not yet verified29,355 works | 30,061 | 60.1% | |
| Could not be verifiedchecked; the source could not be read closely enough | 909 | 1.8% |
Work counts are by a work's best-graded citation. See how support was checked.
Errors found and corrected¶
Reading sources against the text also finds mistakes. When a check establishes one — the source states otherwise, the entry contradicts itself, or the arithmetic does not hold — it is queued for correction. So far 1,132 errors have been corrected in 596 entries, and 87 cited works that no bibliographic authority could find have been withdrawn; others are still waiting.
| Correction | Count |
|---|---|
| Factual errors corrected in the textThe source said otherwise, the entry contradicted itself, or the arithmetic did not hold. | 871 |
| Citations correctedWrong author, year, title or identifier fixed, or a work that did not support the sentence replaced. | 261 |
| Errors corrected, total | 1,132 |
| Works withdrawnNo bibliographic authority could find them, so they no longer count as sources. | 87 |
These are the corrections made so far, not a count of every error the encyclopedia contains: most citations have not been checked yet, and checking continues. Each corrected entry says so in its header and lists its corrections on its version-history page.
What is being cited¶
Kind of source¶
Whether a claim rests on the study that produced a result or on somebody's summary of it is the single most useful thing to know about a bibliography, so it is the first thing shown here.
| Kind of source | Works | Share | % |
|---|---|---|---|
| Primary sourcesWhere a result was first reported — the paper, experiment, dataset, or standard itself. | 14,883 | 42.8% | |
| Secondary sourcesWork that synthesises or teaches rather than reports: reviews, monographs, textbooks, reference works. | 12,840 | 36.9% | |
| Review or monograph | 9,548 | 27.4% | |
| Textbook | 1,973 | 5.7% | |
| Reference work | 1,319 | 3.8% | |
| Grey and web sourcesOutside the formal literature — working papers, institutional reports, web pages. | 5,776 | 16.6% | |
| Grey literature | 2,441 | 7.0% | |
| Web | 3,335 | 9.6% | |
| UnclassifiedThe citation is too thin to classify from. Counted here rather than quietly dropped. | 1,301 | 3.7% |
Where the links point¶
A link's destination is a rough proxy for how durable it is. A DOI resolves for as long as the registrar exists; a publisher's marketing page does not. Grouped by kind of target rather than by hostname, so the long tail of individual publisher domains does not drown the shape.
| Target | Works | Share | % |
|---|---|---|---|
| DOI (doi.org) | 14,981 | 43.0% | |
| Other web | 8,694 | 25.0% | |
| Open Library | 3,653 | 10.5% | |
| Government / official | 1,381 | 4.0% | |
| Internet Archive | 1,028 | 3.0% | |
| Publishers | 607 | 1.7% | |
| Standards bodies | 546 | 1.6% | |
| University presses | 530 | 1.5% | |
| arXiv | 525 | 1.5% | |
| PubMed / NCBI | 354 | 1.0% | |
| Google Books / search | 258 | 0.7% | |
| Wikipedia | 191 | 0.5% | |
| JSTOR | 137 | 0.4% | |
| WorldCat | 89 | 0.3% | |
| Project Gutenberg | 60 | 0.2% | |
| No link yet | 1,766 | 5.1% |
33,034 of the 34,800 works (95%) carry a link; 1,766 do not yet. Nothing is hidden — filter the browse table by link status to see exactly which.
Who published the work¶
Every work with a DOI has a registrant, and the concentration is worth stating: this is the substrate the encyclopedia's citable literature sits on. Attributed from the Crossref record where we hold one, and otherwise from the DOI prefix, which belongs to the registrant.
This covers the 15,595 works we can name a publisher for, not the whole registry — 18,969 works are cited without a DOI at all, and 236 more have one registered by a platform that did not publish them — JSTOR and Project MUSE register DOIs for content they license, so where we hold no record of our own the publisher is left unnamed rather than guessed at from the DOI prefix. Names are curated against Crossref's registrant id rather than matched on spelling, so one house is one row: the twenty-six ways Crossref spells Springer, and ACM's three, are counted together. This table counts registrants, and an imprint counts under its parent — Routledge under Taylor & Francis — with the imprint itself recorded on each work. A press whose DOIs another organisation merely registers keeps its own row: Princeton University Press is not part of De Gruyter's count, and the university presses on JSTOR are not part of JSTOR's.
Publication types¶
Every work carries exactly one publication type. All 23 of them are listed here — the whole vocabulary, not a selection — and every row links into the browse table filtered to that type.
| Type | Works | Share | % |
|---|---|---|---|
| Journal article | 13,693 | 39.3% | |
| Book | 8,792 | 25.3% | |
| Web resource | 3,335 | 9.6% | |
| Book chapter | 1,955 | 5.6% | |
| Report | 1,555 | 4.5% | |
| Unclassified | 1,242 | 3.6% | |
| Conference paper | 971 | 2.8% | |
| Standard | 744 | 2.1% | |
| Preprint | 586 | 1.7% | |
| Reference work | 575 | 1.7% | |
| Course notes rule below | 295 | 0.8% | |
| Legislation | 223 | 0.6% | |
| Court case | 196 | 0.6% | |
| Software rule below | 162 | 0.5% | |
| News article rule below | 140 | 0.4% | |
| Thesis | 110 | 0.3% | |
| Other | 84 | 0.2% | |
| Dataset | 47 | 0.1% | |
| Magazine article rule below | 41 | 0.1% | |
| Artwork rule below | 17 | 0.0% | |
| Letter rule below | 15 | 0.0% | |
| Patent | 12 | 0.0% | |
| Recording rule below | 10 | 0.0% |
Types with a written rule¶
Most of these are self-explanatory. Seven are not, because each borders on a type that already existed and the border is where a vocabulary drifts — so the rule is written down rather than re-derived each time.
Artwork¶
A created work cited as itself: paintings, prints, installations, musical scores, films. Picasso's Le Taureau, Riley's In C, the Eameses' Powers of Ten.
Where the edge is. A document ABOUT the work keeps its own type -- Greenberg's 'Modernist Painting' is a journal article, not an artwork. An exhibition is an event, not a work; cite its catalogue as a book or its archive record as a website.
The corpus cites these as primary evidence of a structure, so a reader needs to know no bibliographic apparatus stands behind them.
Course notes¶
Teaching material the author or their department maintains and hosts: versioned rather than published, unrefereed, revised in place.
Where the edge is. A publisher-issued volume in a series NAMED 'Lecture Notes' is a book. Aczel's Non-Well-Founded Sets (CSLI Lecture Notes 14, ISBN), Lax's Hyperbolic PDEs (Courant Lecture Notes 14, AMS) and Berger & Wolpert (IMS Lecture Notes-Monograph, DOI) are books. Milne's Algebraic Number Theory, Tong's lectures and the MIT OpenCourseWare pages are course-notes. The test is an ISBN or a publisher, not the words in the series title.
The type is named course-notes rather than lecture-notes precisely because 'Lecture Notes' is a book-series name, and using it as the type would invite the confusion this rule exists to prevent.
Letter¶
A letter -- published, archival or open. Pauli's 'Dear Radioactive Ladies and Gentlemen', Maxwell to Tait, Kahneman's open letter to the social-priming field.
Where the edge is. A letter reprinted in a collected-works volume is STILL correspondence; the letter is the form, the book is only its container. A journal named '... Letters' is not correspondence.
In a corpus about abstractions these are primary sources of the first rank -- the neutrino and Maxwell's demon were both proposed in letters.
Magazine article¶
A non-refereed article in a trade, technical, professional or special-interest periodical that has an editorial masthead. Pixel: The Magazine of Scientific Visualization, Technical Analysis of Stocks & Commodities, IEEE Spectrum, Harvard Business Review.
Where the edge is. A NEWS outlet or a general-interest magazine is news-article: the New York Times, the Atlantic, the New Yorker. A PEER-REVIEWED article is journal-article however magazine-like its name. A blog or a vendor documentation page is website.
news-article was added to stop refereed and unrefereed periodicals being conflated, and it stopped one step short. A trade magazine written for practitioners carries different authority from national reporting, and agents classifying the 2026-09-13 campaign had to file the two alike.
News article¶
A piece of journalism or commentary published by a news outlet or a general-interest magazine, whether it appeared in print, online, or both.
Where the edge is. A PEER-REVIEWED article in a scholarly periodical is journal-article, however magazine-like its name. Nature's research papers are journal-articles; Nature's news section is a news-article. A TRADE, technical or professional periodical is magazine-article -- this type is for news outlets and general-interest magazines. A blog, a personal site or a vendor's documentation page is website: the test is an editorial masthead, not the delivery medium.
A reader needs to know whether a claim rests on refereed work or on reporting. Before this type existed the corpus answered that question seven different ways for the same kind of source, and the commonest answer -- journal-article for a Harvard Business Review piece -- was the one that misleads most.
Recording¶
A recording cited as the source: a documentary, a recorded talk or lecture, a podcast or interview episode, a broadcast segment, a conference video.
Where the edge is. A created work cited as itself is artwork -- the Eameses' Powers of Ten is an artwork; a documentary ABOUT the Eameses is audiovisual. The page that HOSTS a recording is not the recording: a channel or series homepage is a website. A transcript published as an article keeps the article's type.
A reader needs to know a claim rests on something spoken rather than written and reviewed. Without the type the corpus scattered these across website, news-article, artwork and null, and agents classifying the 2026-09-13 campaign named it the largest single gap in the vocabulary.
Software¶
The executable artifact itself, or its official documentation, cited as the thing rather than as a paper about the thing.
Where the edge is. A SPECIFICATION of a protocol or format is standard, even when it lives on github.com. The OCI Runtime Specification and the Reactive Streams Specification are standards; SciPy's reference pages, PyTorch's autograd docs, the CUDD package and git are software.
A reader asking what a claim rests on needs to know it rests on an implementation's behaviour, not on a peer-reviewed result.