How references were verified

← Back to references

What was done

Every citation in this encyclopedia went through an adversarial substantiation pipeline. A source was checked twice over: that it exists, and that it supports the specific sentence it is attached to. Those are different questions, and the second is the one that matters. When the pre-existing corpus was first support-checked rather than merely existence-checked, roughly 8% of its citations turned out not to back the claim they sat on — wrong paper, unfilled placeholder, or an annotation copied from an unrelated entry.

Verification was deliberately split into separate passes — find the claims that need a citation, supply one, then confirm it — because combining them degrades every one of the three judgements. A separate layer of reviewers then re-checked a sample from scratch, without seeing who produced the citation.

What this does not establish

A verified citation means a real source supports that sentence. It does not mean the source is correct, that the literature agrees with it, or that the surrounding argument is sound. Where authoritative sources genuinely disagree, the claim was kept and the disagreement noted rather than being quietly resolved in one direction.

The registry, and what it exposes

These 28,668 works were deduplicated from 38,352 citations. Deduplication is not cosmetic: before it, the same book could appear a dozen different ways, under different author renderings, pointing at five different URLs. The registry makes a correction propagate everywhere at once.

It also makes the gaps countable, which is the point:

• 8,961 works (31%) have no authoritative link yet. Many are pre-internet or out-of-print sources; the rest are simply unfinished work.
• 7,826 works (27%) could not be classified by publication type, because their citation is too thin to classify from.
• Editions are counted separately. A 1988 book and its 2013 revised edition are two works, because they are.

What the publication types mean

Every work carries a publication type, and the type facets on the browse pages filter by it. Most types are self-explanatory. Four 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.

correspondence

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.

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.

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.

Contested claims

Claims where authoritative sources legitimately disagree are retained by design. They are not yet individually flagged on the page where they appear. That work is planned and not done, and this page will say so until it is.