last_edit.md — session change log

Running log of edits made by the AI assistant in this working session. Updated: 2026-07-27 (DANAM L1 importer). Newest entries at the top of each section. Legend: ✅ verified (tests/audit/curl pass) · 🆕 new file · ✏️ modified · 🗂️ data/runtime (not a file edit)

0. Status snapshot (current)

DANAM L1 importer: manage.py import_danam_nq ✅ dry-run 25 structures; apply 5 structures + 5 assertions idempotent; parser tests 6/6
Entity resolution: contribution-time linking + refresh_identity_candidates --auto-merge on deploy (§22–23) ✅
Backend test suite: 63/63 green ✅ (plus new test_danam_import)
Nature-rigor audit (kg_rigor_audit): 8/8 invariants PASS ✅
Frontend: 0 TypeScript errors (was 24) · 0 ESLint errors ✅
Backend lint: 146 ruff auto-fixed (import ordering); ~1071 remain (mostly E501 style, F405 settings star-import) — non-gating
Frontend routes swept: 82, 0 broken · Backend API endpoints swept: 308, 0 broken ✅
Services (dev): frontend :3000, backend :8000, Oxigraph container :7879 up
⚠ Parallel edits by user/linter also touched: apps/graph/views.py, apps/cidoc_data/publication_policy.py, src/lib/kg-graph.ts, migration cidoc_data/0016_backfill_metadata_status — noted where relevant.

DANAM corpus L1 materializer (2026-07-27)

🆕 apps/cidoc_data/danam_import/ — NQ parser + structures/assertions materializer
🆕 manage.py import_danam_nq — --dry-run / --limit / --pass / --rebuild / --expected-sha256 / --report-json
🆕 LodExternalIdentity + migration 0021 — idempotent external IRI → row map
✏️ documentation/research/DANAM_CORPUS_INTEGRATION_REPORT.md — §10 usage
🗂️ Smoke: 5 structures + 5 beliefs applied locally; re-run → unchanged skips

Nature-grade KG rigor pass (2026-07-27)

🆕 manage.py corpus_fingerprint + reject-predicate audit (34 preds / 268 quads L0-only)
🆕 manage.py rdf_load_imported_nq (L0; refuses PUBLIC)
🆕 documentation/research/NATURE_KG_RIGOR.md + competency_queries.sparql
🆕 danam_import/licenses.py license matrix; Methods FAIR/CARE + matrix UI
✏️ kg_rigor_audit — L0 isolation HARD + DANAM soft probes
✏️ quality.dangling_edges — exclude skos:exactMatch linkset targets (Nature-correct)
✅ kg_rigor_audit 5/5 HARD + 6/6 soft PASS after fix

1. Deploy bootstrap & RDF triplestore

✏️ entrypoint.sh — added, in order: relationship-predicate seed (--prune); Oxigraph readiness wait + rdf_load_tbox + rdf_rebuild --if-empty; identity bootstrap (bootstrap_identity_clusters, refresh_identity_candidates); backfill_assertion_provenance; kg_rigor_audit (advisory). All idempotent + non-fatal. ✅
✏️ apps/cidoc_data/management/commands/rdf_rebuild.py — new --if-empty flag (rebuild only when public graph empty; safe per-boot). ✅

2. Contribution form ↔ ontology alignment

✏️ ontology/HeritageGraph.yaml — (a) restored CulturalEntity class + category/description slots + CulturalEntityCategoryEnum; (b) added 33 CIDOC-CRM/LinkedArt interop classes + la: prefix (for LUX alignment; non-contributable). ✅
✏️ tools/contribute-hub.yaml — fixed dangling registryKey: cultural_entity → entity (intent + quickStart). ✅
✏️ tools/ui-vizmap.yaml — added HumanMadeObject, Group, Set viz node-types. ✅
✏️ regenerated artifacts (all pass --check): registry.generated.json/.ts, __generated__/heritage-viz-config.ts, __generated__/enums.ts, __generated__/ontology-graph.ts, apps/graph/ontology_config.py, ontology/shapes/generated-heritagegraph-minimal-shacl.ttl, ontology/lod/skos-vocabularies.ttl, ontology/heritagegraph-crm-bridge.ttl, apps/cidoc_data/serializers.generated.py. ✅

3. Contribution → KG pipeline fixes

✏️ apps/heritage_data/serializers.py — fixed CulturalEntityCreateSerializer.create duplicate contributor kwarg (was 500-ing). ✅
✏️ apps/heritage_data/signals.py — sync_cultural_entity_to_public_graph: project to public graph on accept/merge, remove on reject/superseded. ✅
✏️ apps/graph/kg_engine/projector.py — map Cultural-Entity category → museum-renderable CRM class IRI. ✅
✏️ apps/cidoc_data/tests.py — fixed stale OWL_SAME_AS_URI → EXTERNAL_MATCH_URI import (un-blocked the module); updated the rdf_signals test to assert the publication-gated projection (unpublished→not projected, published→projected). ✅

4. Edge projection / store read-coherence

✏️ apps/graph/kg_engine/store.py — (a) _open_local_store_readonly reuses the process's RW handle so reads see their own writes (fixed edges/deletes invisible in embedded/dev); (b) _triple_to_quad handles the @lang convention as a language tag (not a datatype IRI) — fixed crash on language-tagged labels. ✅
✏️ apps/cidoc_data/test_e2e_pipeline.py — wrapped assertion creation in captureOnCommitCallbacks (edge projection is on_commit). ✅

5. Curation gate (publish only reviewed) — partly co-edited w/ user

✏️ apps/graph/views.py — default kg/graph scope to reviewed; coerce anon scope=all → reviewed; keep lux_sampled nodes through the prune step. (User/linter later refactored the gate into apps/cidoc_data/publication_policy.py.)
✏️ src/lib/kg-graph.ts — fetchKgGraph default scope reviewed (also co-edited by user: sourceLayer, includeLux).

6. Yale LUX linked layer (linkset model)

✏️ apps/graph/kg_engine/lux_museum.py — added a bounded connected-sample path (museum_lux_sample_limit, lux_connected_sample_query, batched node fetch at 30/req to avoid GET-URL overflow) so LUX shows as a tagged external layer. ✅
✏️ apps/graph/views.py — surface/keep lux_sampled nodes (co-edited w/ user).
🗂️ Loaded 3.5M Yale LUX quads into the local Oxigraph container (volume hg-oxigraph-data), 3 named graphs; materialized intra-LUX edges; then re-architected to linkset (purged LUX from graph/public). Runtime/data only — not committed.

7. Nature-readiness: evaluation, release, audit

🆕 apps/graph/management/commands/kg_evaluate.py — precision/recall/F1 vs gold standard (types, triples, alignment). ✅
🆕 apps/graph/test_kg_evaluate.py — unit tests for P/R/F1 scoring. ✅
🆕 apps/graph/management/commands/kg_rigor_audit.py — Nature-rigor integrity gate (4 hard + 4 soft invariants; --strict). ✅
✏️ apps/graph/management/commands/kg_purge_public_imports.py — hardened: removal now matches detection (deletes imported-IRI triples it reports). ✅
✏️ apps/graph/test_lux_museum.py — fixed mock for the new sample path + added a sample-path test. ✅
✏️ evaluation/README.md — documented kg_evaluate.
🆕 CHANGELOG.md, 🆕 .zenodo.json, ✏️ CITATION.cff (version 0.1.0, date, DOI placeholder).

✏️ apps/heritage_data/management/commands/seed_relationship_predicates.py — rewrote ontology-driven (88 object-property slots, labels from viz_predicates, --prune). Was 7 hardcoded; relationship-proposal dropdown was empty → fixed. ✅
🗂️ Ran bootstrap_identity_clusters (207 clusters) — identity queue was dormant, now live. ✅

9. Broken links + data hygiene + provenance

🆕 src/app/terms/page.tsx, 🆕 src/app/privacy/page.tsx — were broken links from the public QR-scan consent text; created (content grounded in CC-BY-4.0 + review workflow). ✅
✏️ settings/base.py — RDF_DEFAULT_LANGUAGE = "ne-Latn" (language-tag labels → datatype hygiene). ✅
🆕 apps/cidoc_data/management/commands/backfill_assertion_provenance.py — honest agent+source backfill on accepted edges (provenance coverage 30.5% → 100%). ✅
✏️ apps/cidoc_data/management/commands/seed_test_relationships.py — set attributed_to_agent + source_citation on seeded edges. ✅

10. Known remaining (NOT done — data/scholarly, not engineering)

Dataset scale (small curated corpus); expand the gold standard (1 placeholder entity); novelty/positioning write-up; mint the Zenodo DOI; external-identifier linking (Wikidata/Getty) to raise external_identifier_coverage from 0.
Backend lint debt (~1.2k ruff, mostly E501 — non-gating; CI runs tests not ruff); 24 frontend TS errors (tsc --noEmit) not yet addressed.

11. Memory files updated (`~/.claude/.../memory/`)

project_museum_live_kg.md, project_contribution_form_ontology.md, MEMORY.md — kept in sync with the above findings.

12. Proposed improvements (not yet done) — 2026-06-10

Grounded in inspection this session. Priority: 🔴 high · 🟠 medium · 🟡 nice-to-have.

Frontend / UX

🔴 Museum: LUX-layer toggle + legend. LUX nodes are tagged (sourceLayer:'lux', "Yale LUX" keyFact) but there is no visible show/hide control in FilterBar/museum-toolbar, and no distinct visual style. Add a "Linked (Yale LUX)" toggle + a dimmed/dashed node style + external badge so curated-vs-external is visible, not just in data.
🔴 Fix 24 TypeScript errors (tsc --noEmit): 8×TS2352 (unsafe casts), 6×TS2322 (assignment mismatch), 2×TS2300 (duplicate identifiers), 2×TS18048 (possibly-undefined), etc. Files incl. next.config.ts, access-denied.tsx, entity-qr-code.tsx, heritage-table.tsx, nav-user.tsx, project-add-panel.tsx, reaction-buttons.tsx.
🟠 Empty-state polish: Identity Queue (0 candidates), sparse live curated view, empty projects — friendly guidance instead of blank.
🟠 Accessibility pass: alt text on <img> (incl. entity-qr-code, XR scenes), aria-labels on icon buttons, keyboard nav, contrast — matters for an inclusive heritage platform + journal supplement.
🟡 i18n coverage: next-intl is present; ensure Nepali/Newari locale strings exist (aligns with the ne-Latn data).

Code quality / tests

🟠 Backend lint cleanup: ~1.2k ruff (147 auto-fixable) → ruff check --fix + manual E501.
🟠 Expand test coverage (63 is light): authenticated write flows (contribute submit, review→accept→publish), LUX sample path edge cases, rigor invariants on a fixture.

CI / automation (currently only ontology drift is gated)

🔴 Add CI gates to backend-tests.yml (or new workflow): make check (ontology), ruff check, tsc --noEmit, and a seeded kg_rigor_audit --strict job. Today only ontology-registry.yml runs make check.
🟡 Pre-commit: ensure .pre-commit-config.yaml hooks (ruff, codespell) actually run.

Rigor / data (the publishability path — mostly scholarly)

🔴 Wikidata/Getty reconciliation connector → raise external_identifier_coverage from 0 (the real FAIR-linking contribution).
🔴 Expand gold standard + run kg_evaluate for real P/R/F1 + inter-annotator agreement.
🟠 SHACL strict-on-write (currently validate, non-strict) once shapes verified against the corpus.
🟠 Temporal/EDTF capture (0% events dated) — contribution form field + projection.

FAIR / release

🔴 Mint Zenodo DOI + tag a versioned release (scaffolding already in .zenodo.json/CITATION.cff).
🟠 Per-graph license + PROV on the LUX import (attribution to Yale LUX terms).

Recommended next batch (engineering, high-leverage): museum LUX toggle+legend → fix 24 TS errors → add CI gates → backend lint cleanup.

13. Improvements implemented — 2026-06-10 (from §12)

✅ Fixed all 24 TypeScript errors → 0 (tsc --noEmit clean, ESLint clean):
Session cast (×8 curation pages): as Record<string,unknown> → as unknown as Record<...>.
Component prop mismatches (×5) in knowledge/entity/view/[id]/page-client.tsx: removed invalid entityType from ReactionButtons/ShareButton/ForkButton/EntityComments; added required entityName to ForkButton.
next.config.ts plugin param typed; nav-user.tsx duplicate import removed; entity-qr-code.tsx canvas cast (HTMLCanvasElement); project-add-panel.tsx userDescription ?? ''; reaction-buttons.tsx typed toggleReaction result; heritage-table.tsx schema id → z.union([number,string]); curation/layout.tsx + access-denied.tsx motion {...fadeInUp} → variants={fadeInUp} initial="hidden" animate="show"; curation/contributions/page.tsx token cast.
✅ Backend lint cleanup: ruff check --fix → 146 fixed (import ordering across 14 files); suite re-verified 63/63 green.
✅ 🆕 .github/workflows/frontend-checks.yml — CI gate running tsc --noEmit + ESLint on frontend changes (only ontology-drift was gated before).

Still pending from §12 (next)

🔴 Wikidata/Getty reconciliation · expand gold standard · mint DOI · backend CI ruff/rigor gate (needs lint cleanup / seeded store first) · a11y pass · remaining E501.
(Museum LUX-layer toggle: descoped per user — "no need to show source layer LUX".)

14. UI/UX improvement report — 2026-06-10

🆕 UI_IMPROVEMENT_REPORT.md — code-grounded UX review + prioritized roadmap to increase audience interaction. Top gaps: no global ⌘K search, almost no SEO/OpenGraph (weak shareability), no first-visit onboarding/tour; strong assets (rich viz/XR, social, gamification, AI ChatWidget) under-surfaced.

15. UI/UX improvements — implementation (in progress) — 2026-06-10

✅ Global ⌘K command palette (UX report quick-win #1 — the biggest discovery gap):
npm i cmdk; 🆕 src/components/ui/command.tsx (shadcn command primitive); 🆕 src/components/command-menu.tsx (CommandMenu ⌘K/Ctrl-K dialog with Navigate + Contribute groups, fuzzy filter; CommandMenuTrigger header "Search… ⌘K" button); ✏️ src/app/(dashboard)/layout.tsx mounts both. Verified: tsc 0, eslint 0, dashboard renders 200.
v2 follow-up: extend the palette to entity search over the KG (needs a unified search endpoint + async cmdk).

UX quick wins — ALL 5 shipped & verified (tsc 0, eslint 0, pages render 200)

✅ #0 ⌘K command palette (navigation + Contribute actions) — ui/command.tsx, command-menu.tsx, mounted in layout.
✅ #1 OG/Twitter metadata on public entity pages — generateMetadata in knowledge/entity/view/[id]/page.tsx fetches the entity name server-side (safe fallback) → rich link unfurls. Verified meta tags present in HTML.
✅ #2 First-visit onboarding — 🆕 welcome-dialog.tsx (localStorage-gated, 3 quick links: Museum/Atlas/Contribute, ⌘K tip), mounted in layout.
✅ #3 Surface AI assistant + immersive — added "Ask the AI assistant" to the palette (toggles useChatStore); museum now findable via "xr/immersive/3d" keywords. Fixed a ⌘K collision I'd introduced (ChatWidget also bound ⌘K → removed; chat keeps its FAB).
✅ #4 Re-enabled LanguageSwitcher in the layout header (en + Nepali ne.json locales present and real).
✅ #5 ⌘K entity search over the KG — debounced fetch to /api/v1/cidoc/search/?q=, results grouped → navigate /knowledge/<domain>/view/<id> (plural→singular domain map). Live: "tem" → Taleju Temple etc.
dep added: cmdk.

🧭 Bigger bets remaining (focused passes, multi-week)

narrative story-journeys · collections/bookmarks + profiles/follow + activity feed · mobile-first viz affordances · a11y WCAG AA · full Nepali/Newari UI i18n · embeddable widgets + JSON-LD · dynamic OG images (next/og) · trending/most-viewed (needs view-count telemetry).

16. Professional UI / design-system pass — 2026-06-10 (tsc 0, eslint 0, pages 200)

✅ Typography aligned to design tokens (the #1 pro-UI fix): layout.tsx now loads Poppins (--font-sans) + Fraunces (--font-serif) + Geist Mono (--font-mono) — previously it loaded Geist and mislabeled mono as --font-poppins, so the intended fonts in globals.css never rendered. Body now font-sans.
✅ Discover hub 🆕 src/app/(dashboard)/discover/{page,discover-client}.tsx — hero + Featured + curated-entity grid + "Surprise me"; reuses /kg/graph?scope=reviewed (no new endpoint), parses resource IRIs → /knowledge/<domain>/view/<pk>; uses semantic theme tokens only (no hardcoded colors). Added to sidebar nav + ⌘K palette.

Best-UI-settings recommendations (delivered to user; applied where safe)

✅ Use the design-token fonts (Poppins body + Fraunces display) — done.
✅ Color discipline (global pass) — converted 28 busy gradient-clip static headings → solid semantic tokens across 16 files (home, site-header, leaderboard, about, team, community×2, progression, graphview, curation×5, notification, contribute/pipeline). On-color heroes (graphview/pipeline from-white) → text-white/90; normal headings → font-serif text-foreground. Done via a precise script that only touches classNames with both text-transparent+bg-clip-text. Left intentionally: group-hover: gradient micro-interactions + 1 dynamic per-service decorative gradient (not the anti-pattern). Verified tsc 0, eslint 0, pages 200.
⏳ Consistent page container + section rhythm (shared spacing) — partial; could extract a <PageHeader>/container.
✅ Radius/shadows/tokens well-defined in globals.css — kept; avoid ad-hoc values.
⏳ prefers-reduced-motion for heavy framer-motion; a11y focus-visible + contrast pass.

Next: collections/bookmarks + profiles/follow (bigger bet) — pending.

17. Nature publication — UI assessment & update roadmap — 2026-06-10

Scope: Heritage Museum + cross-app surfaces reviewers will hit (Discover, entity pages, methods/provenance, shareability). Goal: UI credible as a Nature (or Scientific Data / HSS) software or data-descriptor supplement — reproducible, honest about data limits, citable, accessible. No code changes in this pass — assessment + prioritized backlog only (user asked for assessment before edits).

Verdict (honest)

Area	Score	Notes
Viz / interaction	A−	Force graph + map + timeline + XR + story panel is a genuine differentiator
Ontology fidelity (live)	A	Real `rdf:type` + triples via `/kg/graph/` — rare among heritage UIs
Scholarly transparency	C+	Methods popover exists but stale/wrong; edge provenance in API not shown
Reproducibility in UI	D+	No version, DOI, export, or deep-linkable museum state
Accessibility	B−	Good aria on filters/legend/XR; no keyboard graph nav
Publication polish	B	Design-system pass helped; emoji-heavy + inline colors still read “demo” not “journal”

Bottom line: Strong enough to illustrate a paper today; not yet best-in-class for Nature without a scholarly-transparency pass (provenance, citation, export, live-default honesty).

What already works (keep)

Live KG: ontology-typed nodes, real edges, click-to-expand (heritage-museum-client.tsx + kg-graph.ts).
Demo corpus: frozen _provenance, per-image CC attribution (ImageAttribution.tsx, heritage-data.ts).
Methods entry point in toolbar (museum-toolbar.tsx → “Methods” popover).
Ontology mapping block in StoryPanel (HG class, CIDOC-CRM, namespace).
Partial a11y: aria-* on graph SVG, filter bar, timeline; useReducedMotion in XR.
App-wide: ⌘K search, Discover hub, welcome dialog, OG on entity pages, Nepali locale file.

Critical gaps for Nature (🔴 do before submission)

Fix / replace Methods popover copy (museum-toolbar.tsx L141–147)
Still references INSTANCE_CAT_MAP and defensive field extraction — incorrect for live mode (live uses RDF_CLASS_URI_TO_NODE_TYPE + SPARQL projection). Reviewers will notice. Replace with: graph partition IRI, scope=reviewed gate, LinkML/CIDOC-CRM stack, LUX linkset model (curated + skos:exactMatch, not merged dump), SHACL/evaluation commands, link to CITATION.cff / Zenodo.
Surface relationship provenance in the UI
Backend returns provenance on edges (KgGraphEdge in kg-graph.ts); museum drops it when building GraphLink. StoryPanel “Connections” should show source, confidence, asserter, date — required for epistemic claims in a KG paper.
Dataset identity block (every live view)
Show in toolbar or Methods: release version (0.1.0), build/date, named graph URI, entity/edge counts, reviewed scope, license (CC-BY-4.0 data / MIT code), DOI when minted. Link “Cite this software” → CITATION.cff formatted citation.
Demo vs live honesty
Museum defaults to demo (dataSource: 'demo'). For publication deploy: either default live or show a persistent banner: “Illustrative demo corpus — switch to Live for reviewed dataset (N nodes).” Never let reviewers screenshot demo thinking it is the corpus.
Export & reproducibility affordances
Add: Download subgraph JSON / Turtle snippet; copy SPARQL for current view; link to public SPARQL endpoint (RDF_PUBLIC_SPARQL_URL). Nature expects machine-readable access alongside pretty pictures.
Museum node → canonical record
Discover already maps IRIs → /knowledge/<domain>/view/<id> (discover-client.tsx). Museum StoryPanel has no “Open full record” — add it so every node is dereferenceable from the viz.
Deep-linkable museum state
URL query params: ?node=<iri>&view=2d|map|xr&source=live. Enables “see Figure 2” links in the manuscript.

High priority (🟠 before or shortly after submission)

Publication figure mode — high-contrast, reduced chrome, export SVG/PNG (600 dpi) for manuscript figures; optional node labels only, no emoji.
generateMetadata for /heritage-museum — title, description, OG image (static or dynamic graph thumbnail) for supplement links.
Keyboard + screen-reader graph access — roving tabindex on nodes, arrow keys between neighbours, list alternative to force layout (WCAG 2.1 2.1.1 Keyboard).
Confidence / review badges — “Community reviewed” on live entities; dim or style low-confidence edges (use confidenceScore).
Standalone /methods page (not just a popover) — mirrors paper Methods: ontology, pipeline (seed_db → kg_publish → kg_rigor_audit), evaluation (kg_evaluate), LUX attribution (Yale terms), limitations (corpus size, Wikidata coverage).
JSON-LD on public entity pages — schema.org/CreativeWork or Dataset fragment for Google / FAIR discoverability.
Empty / sparse live state — when live graph < 20 nodes, show guided CTA (contribute, run kg_publish) not a bare graph.

Polish (🟡 best-in-class, post-submission)

Replace emoji section headers in StoryPanel with NodeGlyph + semantic tokens (keep emoji optional in “delight” mode).
Remove hardcoded text-blue-400 / inline palette in StoryPanel; use text-primary + tokens only.
Scripted narrative journeys (museum story beats) — aligns with Nature’s preference for guided scientific narrative over raw graph exploration.
Dynamic OG images per entity; embeddable iframe widget for museums.
Full Nepali UI (ne.json exists — audit coverage); Newari if in scope.
Mobile: bottom sheet story panel polish, reduced-node mode on small screens.

Recommended implementation order (engineering)

Phase	Items	Est.
P0 — scholarly credibility	1, 2, 3, 4, 6	2–4 days
P1 — reproducibility	5, 7, 12	2–3 days
P2 — reviewer UX	8, 9, 10, 11, 14	3–5 days
P3 — reach	13, 15–20	ongoing

Cross-reference

Broader UX backlog: UI_IMPROVEMENT_REPORT.md (discovery, social, SEO).
Data/scholarly gaps (not UI): §10 (gold standard, Wikidata, DOI mint).
User note: LUX source-layer toggle descoped (§13) — keep LUX in data layer only unless re-requested.

Recommended next batch for Nature: Methods popover rewrite → edge provenance in StoryPanel → dataset identity + cite block → live-default/banner → “Open full record” on nodes.

18. Nature-rigor museum UI — implemented — 2026-06-10 (tsc 0)

P0 items from §17 shipped in the Heritage Museum + /methods page.

🆕 src/lib/heritage-museum/museum-rigor.ts — citation constants, IRI→detail href, dataset meta from KG response, SPARQL sample, JSON export helpers.
🆕 components/MuseumMethodsPanel.tsx — accurate pipeline copy (LinkML, scope=reviewed, publication policy, provenance edges); dataset identity block; copy citation / SPARQL; link to full methods page.
🆕 src/app/(dashboard)/methods/page.tsx — standalone Methods & data documentation (ontology, pipeline, limitations, LUX attribution).
✏️ museum-toolbar.tsx — Methods popover uses MuseumMethodsPanel; live export JSON button; release badge.
✏️ heritage-museum-client.tsx — defaults to live when API configured; demo banner; sparse-live guidance; preserves edge provenance through GraphLink → attachRelations; dataset meta state; deep-link URL sync (?source=&view=&node=); JSON export.
✏️ StoryPanel.tsx — Open full record link; community reviewed badge (live curated IRIs); per-edge provenance in Connections; text-primary for map link.
✏️ heritage-data.ts — RelationProvenance on relations/links.
✏️ heritage-museum/page.tsx — generateMetadata + Suspense for useSearchParams.
✏️ messages/en.json — methods, export, demo banner, panel provenance strings.
✏️ command-menu.tsx — Methods & data entry.

Still pending from §17 (P2+)

Publication figure export · keyboard graph nav · JSON-LD on entity pages · full Nepali UI strings for new keys · dynamic OG images.

19. Live map + timeline — museum data enrichment — 2026-06-10

Fixed empty map/timeline and awkward 2D layout when using live KG data (coordinates and dates lived on Django rows, not RDF literals).

🆕 apps/graph/kg_engine/museum_graph_enrichment.py — ORM backfill for lat/long/inceptionYear; known Kathmandu Valley place coords; propagate geo along located_at / located_in edges.
✏️ apps/graph/views.py — call enrichment before graph JSON response; expose inceptionYear on nodes.
🆕 apps/graph/test_museum_graph_enrichment.py — unit tests (URI parse, temporal, coords, propagation). ✅
🆕 src/lib/heritage-museum/enrich-museum-graph.ts — client-side geo propagation fallback.
✏️ src/lib/kg-graph.ts — inceptionYear on KgGraphNode.
✏️ src/lib/heritage-museum/temporal-parse.ts — BCE/CE and century strings (e.g. c. 5th century CE).
✏️ heritage-museum-client.tsx — conditional timeline row height; enrichMuseumGraph on live load.
✏️ TimelineStrip.tsx — compact empty state instead of null.
✏️ MapView.tsx — semantic theme tokens + i18n empty state.
✏️ messages/en.json — map.*, timeline.empty* strings.

Verified on live projection: ~33 geo-enriched nodes, ~39 dated nodes after enrichment.

20. XR immersive view — HeritageGraph theme + Nature-rigor — 2026-06-10

Re-themed XR mode from isolated gray-950 / purple WebXR chrome to semantic dashboard tokens (bg-background, bg-card, border-border, text-primary) for light/dark parity with the rest of the museum.

Theme & layout

🆕 src/lib/heritage-museum/xr-theme.ts — shared xrGlassPanel, cinematic overlays, nav chip styles.
✏️ PlaceNav.tsx — card sidebar, search filter (6+ entities), image-first sort, NodeGlyph fallbacks, i18n.
✏️ ImmersiveScene.tsx — full rewrite: theme panels, shadcn Button/Badge, gallery splits with imagery vs text-only (live-data honest), text-first fallback when no hero image, responsive stack (story below identity on narrow screens), parseTemporalAnchor chips (no false “CE” on BCE), Open full record + Community reviewed badges (live), narration/transcript/panorama actions.
✏️ PanoramaViewer.tsx — theme toolbar, i18n, reducedMotion respected in story auto-advance, honest 360° vs flat-photo disclaimers retained, WebXR buttons use secondary/outline not purple/cyan.
✏️ heritage-museum-client.tsx — XR: hide sidebar on <md (gallery picker on mobile), pass dataSource, live badge in XR chrome.
✏️ messages/en.json — heritageMuseum.xr.* (gallery, panorama, WebXR, story controls, search).

Nature-rigor retained / improved

Panorama still labels standard photograph vs true 360° from image aspect ratio (no false VR fidelity).
prefers-reduced-motion: no Ken Burns / parallax / story auto-advance when requested.
Keyboard: arrow keys + +/- zoom in panorama; Escape closes modal.
Image attribution (ImageAttribution) preserved on hero and panorama.

Still pending (XR P2)

Publication figure export from XR frame · scripted narrative journeys tied to ontology periods · true equirectangular assets in curated records (most live nodes are text-only).

21. i18n fix — `heritageMuseum.xr` MISSING_MESSAGE — 2026-06-10

useTranslations('heritageMuseum.xr') failed under next-intl v4 (nested namespace not resolved on client after HMR).

✏️ xr-theme.ts — useXrTranslations(), useMapTranslations(), useTimelineTranslations() resolve via parent heritageMuseum + dotted keys (xr.navTitle, etc.).
✏️ PlaceNav.tsx, ImmersiveScene.tsx, PanoramaViewer.tsx, MapView.tsx, TimelineStrip.tsx — switched to helpers.
✏️ messages/ne.json — added heritageMuseum.xr + map + timeline empty strings (Nepali).

If errors persist after pull: restart the Next dev server (make kill-ports then make frontend) so message bundles reload.

22. Entity resolution — scientific pipeline + form hook — 2026-06-10

Problem

Contributors submitting the same heritage entity twice (e.g. “Pashupatinath” and “Pashupatinath Temple”) each got a separate CIDOC row + singleton EntityCluster, so the live KG showed many apparent duplicates. Identity was bootstrapped 1:1 at deploy but not enforced on new form submissions.

Scientific approach (claim-first ER)

HeritageGraph follows the blocking → matching → decision → human review pattern used in cultural-heritage and LOD entity resolution (cf. spec 005, CIDOC identity layer):

Stage	Rule	Implementation
Block	Never merge across ontology types (`type_scope`)	`location` ≠ `deity` ≠ `guthi`
Match (exact)	Normalized label equality	Auto-link to existing `EntityCluster`
Match (similar)	Distinctive substring (≥8 chars), e.g. `pashupatinath` ⊂ `pashupatinathtemple`	Singleton cluster + `IdentityResolutionCandidate`
Match (none)	No signal	New singleton cluster
Review	Moderator merge/split with audit	`/curation/identity` workspace (existing)
Display	Same cluster → one museum card	`collapseClusterDuplicates()` in live museum (§ prior Pashupati work)

Records stay separate (each submission is a claim); identity.same_referent membership expresses same-referent links without deleting contributor data.

What was implemented

🆕 apps/cidoc_data/identity_label_match.py — normalize_label, labels_are_similar (queue), labels_are_auto_mergeable (unattended merge; no short-token merges like L → Lumbini).
🆕 apps/cidoc_data/contribution_entity_resolution.py — resolve_contribution_identity(); outcomes: linked_existing | singleton_created | candidate_queued | skipped.
✏️ apps/cidoc_data/views.py (ContributionFlowMixin.perform_create) — schedules resolution on_commit after every clusterable CIDOC form create.
✏️ refresh_identity_candidates.py — all clusterable models; exact normalized name and similar-label pairs across clusters.
🆕 run_entity_resolution.py — refresh_identity_candidates + merge_similar_identity_clusters (auto-merge uses strict labels_are_auto_mergeable only).
✏️ merge_similar_identity_clusters.py — switched to labels_are_auto_mergeable; system actor fallback for dev.
✏️ museum_cluster_resolution.py — shared label helpers from identity_label_match.
✏️ heritage_graph/entrypoint.sh (Docker ENTRYPOINT) — bootstrap_identity_clusters + run_entity_resolution on every deploy/restart (idempotent, non-fatal). Root entrypoint.sh also has this for non-Docker use.
✏️ Makefile — make entity-resolution.
🆕 apps/cidoc_data/test_contribution_entity_resolution.py — 6 tests ✅ (RDF_SYNC_ENABLED=false in test env avoids Oxigraph lock).

Contributor flow (form submit)

User POSTs contribute form → new Location / Deity / … row + CulturalEntity review wrapper (unchanged).
resolve_contribution_identity runs on commit:
Exact name (same type) → links to existing cluster; no second cluster.
Similar name → new cluster + row in Identity Queue for moderators.
New name → singleton cluster (same as bootstrap).
After review + publish, museum live view collapses nodes sharing clusterId.

Ops commands

make entity-resolution                    # refresh candidates + safe auto-merge
python manage.py run_entity_resolution --rebuild-rdf   # + rdf_rebuild after bulk dedup
python manage.py refresh_identity_candidates --type-scope=location
python manage.py merge_similar_identity_clusters --label-contains=pashupat --dry-run

Museum UI (live KG duplicates)

Backend: merged clusters + enrich_museum_cluster_identity() attach clusterId / clusterLabel on KG nodes.
Frontend: collapseClusterDuplicates() in heritage-museum-client.tsx shows one card per identity cluster per type; aliases tagged identity:N records merged.

Caveats / next steps

Wikidata / Getty external-ID blocking not yet wired on create (reconciliation API exists on clusters).
Geo proximity blocking for locations — future signal.
Review Identity Queue (/curation/identity) for candidate_queued similar-label pairs; expert curators approve merges.
⚠️ An early dev run with overly loose substring rules may have merged some unrelated rows (e.g. single-char labels); re-run with current code uses MIN_DISTINCTIVE_LEN=8. Inspect queue and split if needed.

23. Consolidation cleanup — redundant files merged/removed — 2026-06-10

Reduced sprawl from incremental museum + entity-resolution work without changing behaviour.

Removed (9 files)

entrypoint.sh (repo root) — duplicate; Docker uses heritage_graph/entrypoint.sh only.
identity_label_match.py — merged into identity_validation.py.
museum_cluster_resolution.py — merged into museum_graph_enrichment.py.
run_entity_resolution.py — thin wrapper; use refresh_identity_candidates --auto-merge instead.
cluster-dedup.ts, enrich-museum-graph.ts, live-node-enrichment.ts — merged into museum-graph.ts.
tools/audit_museum_consistency.py — one-off dev audit script.
UI_IMPROVEMENT_REPORT.md — session artifact; backlog captured in last_edit.md §14–17.

Canonical layout after cleanup

Concern	Single file
Label matching + membership validation	`identity_validation.py`
Form-submit ER	`contribution_entity_resolution.py`
Boot ER pipeline	`refresh_identity_candidates --auto-merge` (+ `merge_similar_identity_clusters` internally)
Museum KG enrichment + cluster metadata	`museum_graph_enrichment.py` + `museum_media.py`
Museum frontend graph helpers	`museum-graph.ts` (+ `museum-rigor.ts`, `xr-theme.ts`, `temporal-parse.ts`)
Docker boot	`heritage_graph/entrypoint.sh` only

Ops (unchanged behaviour)

make identity-candidates   # refresh_identity_candidates --auto-merge
python manage.py refresh_identity_candidates --auto-merge --rebuild-rdf

24. Duplicate contributions — scientific policy implemented — 2026-06-10

Scenario: Contributor A submits basic “Pashupatinath Temple”; Contributor B submits richer details.

Scientific approach (claim-first + evidence-weighted canonical)

Layer	Policy
Prevent	Form duplicate alert while typing (label similarity + type block) → steer to Edit existing
Identity	Same label + type → same `EntityCluster` (no second hub)
Queue	Second submission → `IdentityResolutionCandidate` (`duplicate_contribution_same_cluster`) for curator compare
Canonical	`completeness_score` (narrative, geo, status) picks hub record for museum + hints
Preserve	Both rows kept; reviewer accepts richer claim, rejects or merges thinner duplicate

Implemented

🆕 canonical_record_selection.py — completeness_score, rank_cluster_members, select_canonical_member.
✏️ suggest-duplicates API — label tiers, ranked members, recommendation: edit_existing, registry_key param.
✏️ contribution_entity_resolution.py — exact duplicate links cluster + queues curator review vs canonical member.
✏️ museum_graph_enrichment.py — canonicalMemberId on KG nodes.
🆕 duplicate-contribution-alert.tsx — wired into OntologyForm (all contribute forms).
✏️ museum-graph.ts — collapse prefers evidence-weighted canonical IRI.
🆕 test_canonical_record_selection.py — 7 ER + canonical tests ✅. ```