Wiki · Referencing standard
How Policy Window references
One rule sits under everything: every classification traces to a specific provision of a verified primary source. If a fact cannot be cited to a primary source, it does not publish (charter §7.9(c)). That citation chain is what makes the catalog re-derivablerather than merely asserted — the project's core claim. This page documents the one referencing apparatus every article uses, so a reader (or a reference manager, or an AI assistant) knows exactly what to expect and what a citation here does — and does not — attest.
The apparatus, layer by layer
The same structure renders on every instrument, topic, concept and benchmark article — one source of truth, not four hand-maintained copies. Where the framing legitimately differs (e.g. a concept's “related-topic literature” is not the same as a topic's on-topic corpus), the difference is stated honestly, not flattened away.
1.Provision-level primary citation + excerpt
Each coverage classification (governs / implicit / silent / conflicts) carries a pinpoint to the article or section of the primary instrument, and — where it governs — a short excerpt of the provision itself. Excerpts are shown verbatim in quotation marks, or clearly labelled as a paraphrase when condensed; the two are never conflated. This is the load-bearing layer: the verdict and its evidence sit side by side, both pointing at the source text.
2.One numbered References list (Wikipedia-style)
The analytical prose uses a parsimonious numbered style: in the running text a citation is a small superscript marker (1), so the prose reads cleanly; the marker dereferences to a single numbered References section (#references) at the article foot — one continuous numbering, Wikipedia-style. It holds, in order: the sources cited inline in the analysis (back-linked from their markers, each comprehensive — authors, title, venue, finding, resolver link), then the primary instrument sources behind the classifications as numbered general references. A long DOI/arXiv id never clutters a sentence; whole parenthetical citations (incl. author-year ones that resolve uniquely to a catalogued source) collapse to the bare number, while a date or an un-catalogued aside is left untouched (never turned into a reference). The inline anchors are also exported at /wiki/prose-citations.json.
3.The evidence base (academic & grey literature)
A separate, honestly-framed list of academic and grey-literature sources on the questions the article bears on — not commentary on the instrument itself. Every source carries a source-type label (peer-reviewed, preprint, research institute, civil society, standards, official grey, …), and any one-line finding generated by AI is marked ✦ AI and disclosed as labelled metadata, not an editorial claim (charter §7.9). Browse the whole corpus at the literature index. Academic sources cited in the analysis prose appear as comprehensive numbered notes (layer 2).
4.How to cite — multi-format export
Every article offers a one-click citation in the standard formats reference managers and style guides expect: BibTeX, RIS and CSL-JSON (downloadable, and available as stable routes — /wiki/<slug>/citation.bib, .ris, .csl-json), plus APA, Chicago, Harvard, OSCOLA and Bluebook text. Provision-level citations can also be exported individually.
5.Persistent identifiers + tamper-evident snapshots
Slugs are stable for the life of the archive (renames 308-redirect), and any citation can be pinned to a point in time with ?asOf=YYYY-MM-DD. A snapshot carries a SHA-256 content hash, so a citer can prove later that the cited version was not altered — tamper-evident, not merely tamper-resistant. Full scheme (and the DOI roadmap) at persistent IDs.
6.Verifiability attestation + machine-readable surfaces
You do not have to take the citations on trust. Citation integrity reports — live, computed, never hardcoded — what share of cells carry a primary-source citation, an https source URL, and a provision excerpt, and the catalog ships a content hash so it can be independently re-derived. The whole citation graph is exported open at /wiki/catalog/json, SPARQL, and literature.json.
Benchmarked against Wikipedia
The apparatus is standardised — identically, across the instrument, topic, concept and benchmark article kinds — against Wikipedia's referencing model (MOS:LAYOUT, Citing sources): superscript markers in the running text resolving to one auto-numbered reference list, with a hard line between sources that are cited and material that is merely recommended.
| Wikipedia convention | Policy Window |
|---|---|
| Inline superscript markers → one auto-numbered list | Yes — prose markers (n) resolve to a single numbered References list at the foot of each article, with caret back-links and reused-source dedup. |
| Cited sources in References/Notes | The prose footnotes + the primary instrument sources behind the classifications, in that one list. |
| Further reading — recommended, not cited | The Evidence base (academic & grey literature on the topic) — recommended depth, not inline-cited; positioned as the further-reading equivalent. |
| Consistent headings + shared mechanism on every article | The same shared primitives (one References list, the Evidence base, the cite widget) render identically across the four content article kinds (organisation pages carry the References list only). |
Honest divergences: the end-matter currently lists the Evidence base (further-reading equivalent) beforethe References list, where MOS:LAYOUT orders References before Further reading — the sections are consistent across kinds, but not yet in MOS sequence. And Policy Window adds two things Wikipedia has no equivalent for — provision-level excerpts beside each classification (layer 1) and a machine-readable, multi-format cite-export for the article itself (the “How to cite” widget on each article). Unlike Wikipedia's open editing, every citation traces to a verified primary source under the charter's no-fabrication rule.
What a citation here does NOT attest
- That the cited provision supports the classification. The attestation confirms a citation exists and is well-formed; whether the text actually bears the verdict is editorial judgement, measured separately via inter-rater agreement (Coverage Games).
- That every source URL is live right now. Liveness is checked weekly in CI, not at page load; a source can rot between checks. (Archival capture of cited pages is not yet built — a known gap.)
- Peer-review status. Source-type labels are a structured self-classification, not an external quality score or peer-review certification.
These limits, and the broader ones (instrument-centric ontology, coverage-depth asymmetry), are stated in the methodology §11.