fix(spec): resolve normative contradictions before the v0.1.0 tag

[maehr]

Review findings from #4, fixed before the release is tagged. Everything here is a place where the standard contradicts itself, contradicts the compiler, or documents something that does not exist — cheap now, expensive once v0.1.0 is a citable baseline.

Normative contradictions

• Succession had two mechanisms. specification.md §11 said the prior reference SHOULD be linked to its replacement through an exactMatch MappingAssertion; versioning.md said MappingAssertions MUST NOT be used for succession links. §11 was the stale side — it now points at superseded_by.
• Work.exactMatch / closeMatch vs §6. The compiler emits derived SKOS arrays on Work, but §6 still closed with "External identifiers … are not fields on the Work itself" and omitted both fields from its list. §6 now separates the authoring surface (MappingAssertion, authoritative, carries status and provenance) from the compiler's read-only projection.
• superseded_by was unreachable for deprecated. §11 allowed a deprecated tombstone, §12 attached the successor field only to withdrawn, and the compiler rejected anything but withdrawn/blocked. The guard widens to deprecated/withdrawn/blocked, and §12 now defines the field once, in full.
• A documented compiler invariant did not exist. versioning.md § Compiler invariants claimed a promoted record MUST NOT reference a draft Work/CitationSystem; scripts/compile.ts had no draft check at all (ADR-0003 listed it as an open follow-up). Now implemented for both references and mapping subjects. Moot while all data is draft — live at the first promotion, which is exactly when nothing would have caught it.

enforceTombstoneInvariants is renamed enforceRegistryInvariants since it no longer only handles tombstones, and blocked joins TOMBSTONE_STATUSES so the two guards agree on what a tombstone is.

Published artifacts

api/openapi.yaml never received ADR-0001/0003: it still declared target_kind (the last occurrence in the tree outside CHANGELOG.md and decisions/) and lacked conforms_to, superseded_by, and the Work SKOS arrays. This document is published at /api/.

Housekeeping

• CONTRIBUTING.md claimed vYYYY.MM.N tags are cut from this repo's main, contradicting its own release checklist. This is the unresolved review thread blocking Release v0.1.0: ADR-0001/0002/0003, all data draft, spec fixes #6–#15, Astro 7 (#20) #4 from merging.
• All three ADRs move from Proposed to Accepted — the implementation is merging, so the decision is made. ADR-0002 and ADR-0003 follow-up checklists are reconciled against what actually shipped; sitemap exclusion for drafts is called out as still open rather than checked off, because @astrojs/sitemap is unfiltered and drafts are in sitemap-0.xml.

Verification

npm run format:check, npm run check (0 errors), npm run build (78,940 pages, all internal links valid), npm run compile:data + npm run validate:data (39,250/39,250 records valid).

The new invariant was proved to fire, not just to pass: promoting one work to candidate against the still-draft bekker system produced 5,220 violations and failed the build with promoted reference (candidate) points at draft system https://textrefs.org/id/system/bekker. Reverted afterwards; the data/ pointer is unchanged at c0a3275.

Changelog regenerated with git-cliff --tag v0.1.0 — plain npm run changelog retitles the section [Unreleased], since the release checklist regenerates before the tag exists.

🤖 Generated with Claude Code