data: second-wave work candidates with established citation systems

[maehr]

Affected file, TextRefs ID, or URL

New files under works/ and systems/. No existing record changes.

Citation system

Seven new ones, listed below.

What is wrong or missing?

The registry holds twelve works. The ROADMAP plans to "grow the registry with more canonical
examples across theology, classics, philosophy, law, and other citation traditions."

The corpus survey (tracking issue) sorts its candidates into three waves. This issue is the
second wave: works whose citation system is beyond dispute and stable for a century or
more, but whose online resolver is provider-specific and needs mapping work. The first wave
— the near-automatic reference spaces — is handled in separate issues.

Every row below is unverified. It comes from an LLM-generated survey whose sourcing is
uneven; see the tracking issue. No row is ready to implement. Each needs the treatment #16
and #19 got: open the URLs, test an invalid locator as a control, confirm the resolver
addresses the passage and not just the work.

Suggested correction or new value

Work one row at a time. Open a separate issue per work once it is verified.

• Dante, Divina Commedia — dante.commedia. System: cantica, canto, verse
  (Inf.1.1, Purg.13.25, Par.33.145). The cantica is part of the locator, so the
  system needs a three-part regex, and Inf/Purg/Par must be a fixed vocabulary.
  Resolver candidates: Dartmouth Dante Project, Princeton Dante Project, Dante Lab. The
  Petrocchi text is the critical edition all three use. Open question: Dante Lab's
  reader takes reader[cantica]=2&reader[canto]=5 as query parameters and stops at the
  canto — it does not address the verse. Check whether any of the three addresses a
  single verse.
• Thomas Aquinas, Summa Theologiae — aquinas.summa-theologiae. System: pars,
  quaestio, articulus (ST.I-II.q90.a1). The pars uses Roman numerals with a hyphenated
  form for the Secunda secundae, so the locator vocabulary needs deciding before the
  regex. Resolver candidate: Corpus Thomisticum (Universidad de Navarra), which is the
  standard scholarly Opera omnia. Open question: Corpus Thomisticum has no
  documented resolver API. Check whether its URLs are deterministic per article.
• Justinian, Digest — digest or corpus-iuris-civilis.digest. System: book,
  title, lex, paragraph (D.41.2.3pr). Note pr (principium) is a non-numeric
  paragraph, so the regex must admit it alongside integers. Resolver candidate: the Roman
  Law Library (Université Grenoble Alpes), whose Digest text is Mommsen/Krueger. Open
  question: the site is static HTML with anchors. Confirm the anchors are per-lex and
  stable.
• Shakespeare — one work per play, so this row is really a sub-project. System: act,
  scene, line. Open question, and it is the interesting one: Folger Through Line
  Numbers (FTLN) are Folger-specific, while act/scene/line varies between editorial
  traditions. The survey argues act/scene/line is the abstract identity and FTLN is a
  resolver mapping. That is the right shape for TextRefs, but it needs the provider
  vocabulary mapping (feat(compile): map locator variables into a provider's own vocabulary textrefs.org#71) and an FTLN concordance. Resolver
  candidate: Folger Shakespeare, which documents an API that resolves a single FTLN.
• Nietzsche — one work per title (nietzsche.frohliche-wissenschaft, …). System:
  work sigla plus aphorism number (FW 125). Resolver candidate: Nietzsche Source
  eKGWB, based on Colli/Montinari (KGW/KSA), which is the reference edition. Open
  question: the URI is hash-based (nietzschesource.org/#eKGWB/FW-125). A fragment is
  never sent to the server, so this cannot be confirmed by fetch and needs a browser
  check. Also: the sigla belong in alternative_labels, and the work-per-title split
  means the sigla, not the aphorism, carry the work identity.
• Kant — Akademie-Ausgabe volume and page (AA IV, 421). Open question, and it
  may be disqualifying: an AA citation addresses a page of a specific edition, not a
  structural unit of the work. That is closer to the printed-page citation TextRefs
  exists to replace than to Stephanus or Bekker, which are also page-based but function
  as edition-independent structure. Decide whether AA is a citation system TextRefs
  should carry before modelling it. Resolver candidate: the Kant-Korpus at
  Uni Duisburg-Essen, whose URL pattern is deterministic (zero-padded volume, page).
  The survey's source for this URL pattern is a ZDF fact-check PDF. Re-source it.
• Hume — hume.treatise, hume.enquiry-human-understanding. System: book, part,
  section, paragraph (T 1.3.6.4, EHU 10.1). Resolver candidate: Hume Texts Online
  (Peter Millican, Hertford College Oxford), the reference text the Hume Society
  supports. Open question: the survey says a siglum-to-path table maps EHU 10.1 to
  /texts/e/10 with the paragraph as a DOM anchor. That means a per-work sigla map, not
  one template. Confirm the anchors exist.

Evidence and sources

None of the above is verified. The sources are data/research/a.md and data/research/b.md,
now carried in the tracking issue. Their quality is uneven and two citations in the first
file are unusable. Treat every claim here as a lead, not a finding.

The one thing already learned the hard way, from the Perseus/Scaife companion issue: a
resolver that returns HTTP 200 may still be serving the wrong passage. Always test an
invalid locator as a control before believing a link works.

Rights or licence note

To be established per work. Several candidates are academic sites with no explicit licence,
which is a license_url:-only case at best. Record access conditions during verification.

[maehr]

Part of #28, the corpus survey tracking issue.

[maehr]

Verification pass done — all seven rows checked on 2026-08-25, each valid case paired with an invalid locator as a control, per the rule this issue sets.

Five of the seven are viable. Per-work issues opened as this issue instructs:

Row	Verdict	Issue
Shakespeare	PASS — strongest of the seven	#33
Hume	PASS	#34
Digest	PASS	#35
Nietzsche	PASS	#36
Dante	PASS (Princeton; two candidates rejected)	#37
Aquinas	qualified negative — see below	—
Kant	unreachable today — see below	—

Three of this issue's open questions are answered, and two of them dissolve the row's difficulty.

• Shakespeare needed "an FTLN concordance". It does not. Folger resolves act/scene/line directly: …/Ham/line/3.1.64 → To be or not to be—that is the question:. No FTLN intermediary and no provider-vocabulary mapping, so feat(compile): map locator variables into a provider's own vocabulary textrefs.org#71 is not a dependency for this row. The concordance is available anyway — …/Ham/ftln/1234 publishes Line: 2.2.170 — but nothing needs it.
• Nietzsche's blocker is obsolete. This issue says the hash URI "cannot be confirmed by fetch". Nietzsche Source now serves a server-validated HTTPS path; FW-9999 and XX-1 return real 404s. It is link-checkable.
• Dante: yes, one of the three addresses a verse. Princeton, via INP_START. Confirmed with a cross-control — the Inf 1.1 page contains none of Inf 5.121's text, and the reverse.

---

Aquinas — qualified negative. Do not implement; two concordances are missing, not one.

This issue asks: "Corpus Thomisticum has no documented resolver API. Check whether its URLs are deterministic per article." They are not.

URLs are sth{P}{QQQ}.html, first digit = pars (1=I, 2=I-II, 3=II-II, 4=III), last three = the starting question of a chunk. Chunk size is irregular, so the URL cannot be derived from a question number:

URL	Covers
sth1001	I q.1 (one question)
sth1002	I q.2
sth1010	404
sth1050	I q.50–64 (fifteen)
sth2001	I-II q.1–5 (five)
sth2002	404
sth3001	II-II q.1–16 (sixteen)
sth4001	III q.1

Only the first question of a chunk has a URL, so a question→chunk lookup table over ~512 questions would be needed.

And within a page there is no article anchor. Anchors are opaque sequential global IDs — <A NAME="28232">, 66 of them in I q.1 — a paragraph counter across the whole Opera omnia, unrelated to pars/quaestio/articulus. "Articulus 1…6" appears as plain text with nothing addressable. So ST.I-II.q90.a1 cannot be reached at all; the best achievable is a chunk landing page, which for sth3001 means sixteen questions resolve to one URL.

Controls are well behaved: sth1010, sth9999 and opera.html all return a real HTTP 404.

Alternative checked and dead: aquinas.cc returns HTTP 404 with a 0-byte body on every path, including the root.

Same class of blocker as the Scaife concordance in #22 — a large data-entry job for one link. Defer.

Kant — unreachable on 2026-08-25. Not verifiable today, and possibly the wrong question.

korpora.zim.uni-duisburg-essen.de resolves to 132.252.238.56, but the TCP connection never completes. Two independent clients, both schemes:

• curl over https and http → Connection timed out after 20004 ms, HTTP 000
• Chrome → navigation fails, tab reverts to the new-tab page

So the URL pattern this issue asks to re-source cannot be re-sourced today. Please do not record the Kant-Korpus as dead on this evidence — a university host may be temporarily down or firewalled, and this is one day's reading.

The prior question in this issue matters more anyway, and is answerable without the site: whether an Akademie-Ausgabe volume-and-page citation is a citation system TextRefs should carry, given that it addresses a page of a specific edition rather than a structural unit. Settle that first; it may make the resolver moot. Lowest priority of the seven.

[maehr]

Decision on the open rows. The rule: we only carry citations that are canonical and commonly used, and that have an authoritative way to resolve online. If either test fails, the work is not added — it is not parked as "blocked".

Applying it to the two rows that produced no issue:

Aquinas — closed, will not add

ST I-II q.90 a.1 passes the first test easily. It fails the second. Corpus Thomisticum's page chunks are irregular (sth1001 = one question, sth2001 = five, sth3001 = sixteen, sth1010 = 404), so question→URL needs a hand-built lookup over ~512 questions; and its in-page anchors are opaque global counters (<A NAME="28232">) unrelated to the article, so the articulus is unreachable at any price short of a second concordance over ~2,700 articles. aquinas.cc is dead. There is no authoritative online route to the cited passage, so there is nothing to add.

Kant — closed, will not add

Fails both tests, which is why it was always the weakest row.

AA IV, 421 addresses a page of one specific edition, not a structural unit of the work. That is the printed-page citation TextRefs exists to replace, not an edition-independent structure like Stephanus or Bekker — this issue raised that possibility itself and called it "may be disqualifying". It is. And separately, the Kant-Korpus host would not accept a TCP connection from either curl or Chrome on 2026-08-25, so there is no verified resolver either.

The other five stand, and three are now implemented

Row	Status
Hume	implemented — PR #38, 255 + 1,213 references
Dante	implemented — PR #39, 14,233 references
Nietzsche	verified, #36 — canonical sigla, authoritative provider (Colli/Montinari), server-validated URLs
Digest	verified, #35 — canonical, authoritative (Mommsen/Krueger), anchors are the citation itself
Shakespeare	verified, #33 — canonical act/scene/line, authoritative (Folger)

All three remaining rows pass both tests, so they are additions waiting on data entry rather than open questions.

Suggest this issue now tracks only those three and closes when they land. The Aquinas and Kant rows are settled and should not be reopened without new evidence — specifically, a per-article resolver for Corpus Thomisticum, or a decision that Akademie pagination is a citation system TextRefs wants.