feat(works)!: resolver review — drop non-resolving targets, add NA28/BHS/SuttaCentral/NINJAL

[maehr]

Closes #15, #16, #17, #18, #19

Implements a scholarly review of all 12 work records, comparing the configured resolver targets against the editions each field actually cites. Every claim below was re-checked against the live sites on 2026-08-12; three of the review's own claims turned out to be wrong and are corrected here rather than acted on.

Everything is draft under ADR-0004 and no resolver target has ever been promised, so entries that do not work are deleted rather than demoted. No resolver-quality vocabulary is introduced.

Deleted — six resolvers that do not address the cited passage (#15)

Four had no locator variable in the URL at all, so every reference in the work expanded to the same landing page:

File	Resolver	References sharing one URL
aristotle.nicomachean-ethics.yaml	Project Gutenberg 8438	~5,220
plato.republic.yaml	Project Gutenberg 1497	1,475
murasaki-shikibu.genji.yaml	Project Gutenberg 19264	54
dhammapada.yaml	palikanon.com	423

That is worse than no resolver: the record asserts "here is Republic 514a" and hands the reader the front matter of an ebook. PG 19264 is additionally not the work — it is Wilson's anthology Japanese Literature, carrying only Suematsu's selections.

Plus:

• new-testament.yaml's Wikisource KJV entry, which is broken rather than coarse: en.wikisource.org migrated most books to one continuous page each, so .../{Book}/Chapter_{n} 404s for 26 of the 27 NT books, and where the pages survive the #{verse} anchors are not unique (Genesis carries 15 separate id="1" anchors; Romans has none).
• dhammapada.yaml's Project Gutenberg entry, which does anchor per chapter but is an access copy, not a critical edition — Gutenberg's own mission statement disclaims establishing textual standards.

Wikisource is kept everywhere it uses a real url_by map that reaches the cited chapter (Analects, Daodejing, Genji), as is ancient-buddhist-texts.net.

Added — the four missing editions

SuttaCentral, Dhammapada (#16). Pali root text and Sujato translation, CC0, at verse granularity: 26.41 → https://suttacentral.net/dhp423/en/sujato. The compiler already derives {verseGlobal} from the system's chapter_sizes, so this needed no new machinery. Turns 423 chapter-level targets into 423 verse-level ones.

NINJAL, Genji (#17). The Library of Congress manuscript transcription, CC BY 4.0, all 54 chapters via url_by. The file's own comment had flagged this as pending.

NA28, New Testament (#18) and BHS, Tanakh (#19). Both on die-bibel.de, both free to read without registration, both with per-verse anchors:

John.3.16 → https://www.die-bibel.de/bibel/NA28/JHN.3/#JHN.3.16
Ps.23.1   → https://www.die-bibel.de/bibel/BHS/PSA.23/#PSA.23.1

Both use the new vars: mapping (textrefs/textrefs.org#72) because die-bibel.de addresses books by USFM code and the locator carries OSIS codes. Neither is openly licensed, so access: open with license_url: only and no SPDX license:.

Completed — the two Bible works were 90% missing

Both records were named for a whole corpus while enumerating a single book:

Work	Before	After
New Testament	John only, 879 refs	27 books, 7,953 refs
Tanakh	Genesis only, 1,533 refs	39 books, 23,213 refs

Counts derived mechanically, not transcribed:

• NT from morphgnt/sblgnt — the same edition the STEP Bible resolver serves.
• Tanakh from openscriptures/morphhb wlc/*.xml, the Westminster Leningrad Codex, i.e. the Masoretic text BHS presents. Its 39 files are already named with OSIS codes, so the vocabularies line up with no translation step. 23,213 matches the standard WLC figure.

Each entry is the highest verse number in the chapter, not the number of verses printed. SBLGNT omits verses that traditional numbering reserves (Matt 17:21, John 5:4, the bracketed pericope adulterae) and those numbers are still cited; a citation registry should mint Matt.17.21 even where a critical text prints nothing there.

Thirty-nine books rather than the twenty-four of the Jewish reckoning, because the locator vocabulary is OSIS, which splits Samuel/Kings/Chronicles and enumerates the Twelve individually.

Three corrections to the review that prompted this

• Sefaria is not broken. The review flagged the Tanakh template as substituting OSIS Gen where Sefaria's URLs read Genesis. Sefaria's ref parser normalises OSIS itself: sefaria.org/Gen.1.1 → 301 → Genesis.1.1 (200), same for Exod.20.1, 1Sam.1.1, Ps.23.1. Left unchanged.
• STEP Bible is not broken. It resolves OSIS codes server-side (John.3.16, Matt.1.1, 1Cor.13.4, Rev.1.1 all return the right SBLGNT text). Left unchanged.
• SHEBANQ was dropped, not added. BHS itself turned out to be freely readable on die-bibel.de, which is what the review actually wanted. SHEBANQ is CC BY-NC, renders verse text only via client-side JS (so per-verse resolution could not be confirmed), and would need a third book vocabulary (Latin/Vulgate: Samuel_I, Psalmi, Canticum) — a second mapping table for no added coverage.

Verification

• npm run verify → clean (format, astro check, 18 tests, full build) in 8m 07s.
• npm run build:data → 67,958 references, 24 mappings, 68,004 records, all valid, zero skipped resolver entries.
• The zero matters: an unmapped book code skips its entry and warns, so a non-zero count would mean a hole in one of the two 27/39-entry USFM tables.
• All 66 USFM codes checked individually against the rendered chapter title. die-bibel.de answers an unrecognised code with HTTP 200 and a default page — BHS/GENE.1/ renders "Genesis 1", and the OSIS spellings MATT.1/JOH.3/1COR.13 all silently render Matthäus 1 — so status codes prove nothing. All 66 titles came back distinct and correct.
• Compiled output spot-checked for dhammapada|26.41, genji|54, John.3.16, 1Cor.13.4, Gen.1.1, Ps.23.1, and asserted to contain zero remaining gutenberg.org, palikanon.com or Wikisource-KJV URLs.

One caveat recorded honestly: SuttaCentral is a client-side SPA that returns 200 for any path, including out-of-range /dhp424. Its per-verse route is evidenced by SuttaCentral's own Feb-2022 release note introducing exactly this linking form, and by the API's range structure — not by DOM inspection. Details in #16.

Sequencing

• Stacked on feat(works)!: reclassify mappings to alternateOf and isReferencedBy (ADR-0006) #13 (ADR-0006 reclassification), since main still carries exactMatch/closeMatch, which the parent's current source-schema.ts rejects outright. Retarget to main once feat(works)!: reclassify mappings to alternateOf and isReferencedBy (ADR-0006) #13 merges.
• Blocked on feat(compile): map locator variables into a provider's own vocabulary (#71) textrefs.org#72, which adds the vars field. Until that reaches staging, this PR's validate job is red for that reason and no other.

[maehr]

CI status, so the red check is not misread:

validate — fail, inherited from #13, not from this change. The failure is byte-identical to the one on #13: mapping/d5d6ca80-…: relation: Invalid option: expected one of "exactMatch"|"closeMatch". Registry CI validates against textrefs.org@staging, which does not yet have ADR-0006, so it rejects alternateOf — and it fails there before ever reaching a resolver, let alone the new vars field.

Two things must land before this can go green, in order:

1. docs(decisions): decide mapping relation vocabulary (ADR-0006) textrefs.org#67 → staging (unblocks the relation vocabulary)
2. feat(compile): map locator variables into a provider's own vocabulary (#71) textrefs.org#72 → staging (adds the vars field this data uses)

Locally, against the correct parent, npm run build:data compiles clean: 67,958 references, 68,004 records, all valid, zero skipped resolver entries.

[Copilot]

Pull request overview

This PR updates multiple work records in the TextRefs registry to reflect a resolver-quality review: removing resolver targets that don’t address the cited passage, adding missing scholarly/field-standard resolver targets, and expanding Bible corpora reference enumerations to full book coverage.

Changes:

• Removed non-resolving/coarse resolver targets (notably Project Gutenberg and other landing-page-only resolvers) and added explanatory rationale where a second resolver is intentionally absent.
• Added new resolvers: SuttaCentral (Dhammapada, verse-level), NINJAL (Genji, per-chapter url_by), and Deutsche Bibelgesellschaft NA28/BHS (via vars book-code mapping).
• Expanded references_range enumerations for New Testament (27 books) and Tanakh (39 books) with chapter-level maxima.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
works/tanakh.yaml	Adds BHS resolver (die-bibel.de) with OSIS→USFM mapping and expands Tanakh references to all 39 books.
works/new-testament.yaml	Adds NA28 resolver (die-bibel.de) with OSIS→USFM mapping, removes broken Wikisource KJV resolver, and expands references to all 27 books.
works/dhammapada.yaml	Adds SuttaCentral verse-level resolvers (Pali + English) and removes non-verse/non-resolving targets.
works/murasaki-shikibu.genji.yaml	Adds NINJAL Library of Congress manuscript transcription resolver via per-chapter url_by mapping and removes Project Gutenberg.
works/plato.republic.yaml	Removes Project Gutenberg landing-page resolver; keeps Perseus with rationale for no second resolver.
works/aristotle.nicomachean-ethics.yaml	Removes Project Gutenberg landing-page resolver; keeps Perseus with rationale for no second resolver.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.