docs: publish an ontology stub at https://textrefs.org/ontology#

[maehr]

Context

public/contexts/v1.jsonld binds the tr: prefix to https://textrefs.org/ontology# and then uses it for 20 terms: four classes (tr:Work, tr:CitationSystem, tr:CanonicalReference, tr:MappingAssertion) and sixteen properties (tr:key, tr:locator, tr:relation, tr:status, tr:workKey, tr:citationSystemKey, tr:preferredCitationSystemKey, tr:creatorKind, tr:subject, tr:target, tr:identifier, tr:resolverTargets, tr:access, tr:lastChecked, tr:locatorRegex).

None of them dereferences. https://textrefs.org/ontology# returns nothing. Every published record therefore carries twenty terms that a Linked Data client cannot resolve to a definition.

ADR-0006 named this and declined to fix it there:

Publish an ontology stub at https://textrefs.org/ontology#. Pre-existing debt (tr:Work, tr:relation, tr:locator are all undereferenceable), not created here, but option 4 above was rejected partly on its account.

That last clause matters. ADR-0006 chose prov:alternateOf and dcterms:isReferencedBy over minting tr: terms partly because a tr: term would have been one more undereferenceable IRI. The debt is now shaping decisions, not just sitting behind them.

What the stub needs

1. A document served at https://textrefs.org/ontology, with the fragment identifiers the context uses.
2. One definition per term: a label, a comment, and — where it is honest to state one — a domain and a range. A stub may leave a term loosely specified. It may not leave it undefined.
3. Content negotiation is not available. The site is static and GitHub Pages derives every Content-Type from the file extension, as api/openapi.yaml already documents. Publish HTML with embedded RDFa or JSON-LD, or publish a Turtle file at a distinct path and link it, rather than promising negotiation the host cannot perform.
4. Versioning. The context is v1. The ontology needs a stated relationship to it — whether the terms are versioned with the context or independently.

Decide first

• Reuse where a term already exists. Several tr: properties may be expressible with SKOS, DCTerms or schema.org terms that do dereference. A stub is the moment to check, because replacing a term after v0.1.0 is a context change with published consumers.
• Scope. All 20 terms, or the four classes first?

Links

• decisions/ADR-0006-mapping-relation-vocabulary.md — the follow-up and the rejected option 4
• public/contexts/v1.jsonld — the 20 terms
• src/content/docs/standard/json-ld.md — the published context documentation

[maehr]

Fixed on staging by 2d71119 (#127). All 17 tr: terms are defined in src/lib/ontology.ts and served at /ontology/ (HTML) and /ontology.json (JSON-LD). Three of the terms this issue counted are gone rather than defined: key is now dcterms:identifier and locator is skos:notation, reusing an established term where one fits, as the issue asked.

The fix ships with the v0.1.0 release, #4. Closing now rather than at the tag, because the change is on staging and Closes does not fire from a merge into a non-default branch.