Improvement intake: public representation repair

Report body

Improvement intake: public representation repair

Work date: 2026-09-08. Starting checkpoint: c9cc470156e28e4f61e291c369d41c6e20ba5ba9. This is current implementation evidence, not a claim that the whole intake is closed.

The user's correction is accepted: reports still in Improvement are unfinished, and recording an inventory or full review is not enough to call them processed. Close only actual selected work; preserve rejected suggestions and exact sources. General release/independent-adoption gates are tracked separately from the work actually accepted from an individual report. They must not hide file-specific unfinished work or keep already disposed sources parked forever.

Concrete defects confirmed

The current glossary contained two identical DefinedTermSet objects and zero hasDefinedTerm entries despite 120 visible terms. The current root graph had WebSite and WebPage identities but did not resolve the #organization publisher identity referenced by existing documents. A parity test additionally found DEF067's source name still said Cross-Project Contradiction Register while the visible glossary already used Cross-Project Clarification & Correction Register.

These are the concrete remaining structured-data gaps identified in reports/v048-agent-integration-intake-review.md, not speculative recommendations or a general SEO score.

Implemented owners

• scripts/refresh_public_semantics.py derives one glossary term set from data/_source/definitions.json, adds stable visible row anchors, refuses silent source/visible drift and preserves existing page content and brand.
• The 120 descriptions and labels are identical in the registry, visible table and machine term entries. DEF067 now uses its already-published name; the definition itself is unchanged. The existing data/definitions wrapper is regenerated from the same source, preserving API/source parity.
• data/_source/publisher-identity.json owns the narrow first-party publisher declaration. The homepage resolves #organization without invented corporate registration, personnel, dates, partner affiliations or delegated authority. A visible footer statement preserves referenced research authorship/authority.
• scripts/test_public_semantics.py guards unique/full term sets, real anchors, exact descriptions, idempotence, drift rejection, duplicate-ID rejection, safe script serialization, publisher scope and protected hero references.

Primary vocabulary references fetched during this work: DefinedTermSet and publisher. These support term membership and the publisher representation, not search indexing, ranking, citation, adoption or institutional authority. No such outcomes are claimed.

Verification state

Nine focused tests passed. Generator execution passed and a second --check reported no stale output. The initial test correctly detected DEF067 drift; the source was reconciled to the existing visible wording before generation. Browser inspection used the local static preview in the in-app Chromium browser. The homepage desktop hero and publisher footer retained the existing visual identity. Keyboard focus on the skip link had a visible solid outline and the link navigated to #main-content. At 390px with page scripts disabled and reduced motion emulated, the homepage remained readable without document-wide overflow.

The inspection exposed a real glossary defect: narrow columns broke ordinary words into near-vertical text. assets/glossary.css and the existing projection now give the table a readable intrinsic width inside a named, keyboard-focusable scroll region with a caption and a visible scrolling explanation. This is a page-scoped addition, not a shell or brand rewrite. At 390px and 320px, the table had 768px scroll width while document scroll width stayed inside the viewport. Arrow-key input changed the scroll region's horizontal position and focus had a solid outline. Fresh navigation to #def067 positioned its row below the sticky header (245px versus a 203px header bottom). Reload scroll restoration is not treated as a new-fragment test. All 120 stable row IDs and the corrected DEF067 label were observed in the DOM. Temporary browser emulation was reset afterward.

These are bounded browser observations, not a complete accessibility standard, performance benchmark, production deployment or full-intake completion claim. Production readback subsequently passed in reports/v049-replay-live-evidence.json: the homepage, glossary, scoped CSS, definitions endpoint and updated agent guide matched the committed cc36d0821a7f9940fd0768395918afdb300e4894 root package exactly. The deployment is a six-file patch with rollback custody, not a claim that the whole live tree or historical research glossary was rewritten to the new version.

The production in-app browser check observed all 120 row anchors, readable table columns, and #def067 below the sticky header at 390px without page JavaScript and with reduced motion enabled. Table client/scroll widths were 342/768px; document width was 375px within a 390px viewport. At 320px the document was 305px while the table stayed locally scrollable. Arrow-key input moved the table and its focused region had a solid outline. Homepage document widths stayed inside 390, 768, 1024, 1366 and 1920px viewports; its protected image loaded at 1672x941 and remained visually intact. Keyboard Tab reached the visible skip link; activating it and pressing Tab moved into main content to Enter coordination rooms with visible focus. Temporary script/motion/viewport overrides were reset and the temporary browser tab closed. These checks do not assert high-zoom, all-assistive-technology, full WCAG or Core Web Vitals certification.

Preserved scope

Unread reports and unrelated untracked machine-controller drafts are untouched. No scheduled task was created; the prior recurring-write rejection still stands until explicitly authorized. Optional parallel review was requested separately, but no review agents were spawned before the user's response.