Cross-domain links: registered offices → municipalities, counties → NUTS3, MPs → constituencies

[namedgraph]

Summary

Connects the three previously-disjoint entity islands (admin units, legal entities, Seimas) with deterministic, re-runnable ETL that pulls the join keys from live open-data sources — no hand-authored CSVs. Closes three items from the ONTOLOGY-NOTES backlog (registered office → municipality, NUTS/LAU exactMatch, Seimas constituency).

Legal entity → administrative unit (registered office)

JuridinisAsmuo carries no usable address, so the office resolves JAR Buveine (ja_kodas → aob_kodas) ⋈ AR Pastatas (aob_kodas → savivaldybe.sav_kodas + pasto_kodas) on aob_kodas, keeping the latest office. Modelled as the SEMIC chain org:hasRegisteredSite → org:Site → org:siteAddress → locn:Address, with locn:adminUnit → admin-units/{sav}/#this (committed primary) and locn:postCode. Postcode is an attribute, not a join key — the exact aob_kodas → sav_kodas beats the not-guaranteed-1:1 postcode→municipality. Adds Buveine/Pastatas fetches + institution-scoped awk narrowing, org:Site/locn:Address SHACL shapes, and the :OrganizationsHere inverse view.

County → NUTS3

reconcile.py emits skos:exactMatch → data.europa.eu/nuts/code/LT0xx from Wikidata P605 — a couple of extra columns in the existing WDQS enrichment (10 counties; municipalities are LAU, no P605, and inherit NUTS3 via their county).

MP → electoral constituency (apygarda)

Parses the 71 single-member districts from the members feed (isrinkimo_budas) into a new constituencies/{nr}/ container (ltlod:ElectoralDistrict); the single-member seat tenure links via ltlod:electoralDistrict. An apygarda is not an administrative unit — own container, no dct:isPartOf into admin-units/. Party-list ("Pagal sąrašą") seats have none. New container doc, SHACL shape, :MembersElectedHere view.

Verification

• Gap A — mapping tested on a 3-entity fixture (latest-office selection; the no-Pastatas fallback degrades cleanly); SHACL conforms; both Makefiles dry-run clean.
• Gap C — run live: exactly 10 counties get NUTS3 (LT011–LT029), 642 existing matches unchanged.
• Gap B — run on the cached feed: 71 district graphs, 74 MP links (party-list correctly excluded), 0 dangling references, SHACL conforms.
• Both new example queries run correctly on sample data; all changed .ttl/.rq/.py syntax-checked.

Not included

Committed datasets/current/ snapshots are not regenerated in this PR — that mixes in heavy national Buveine/Pastatas fetches (offices), unrelated Wikidata image drift (NUTS), and nondeterministic arq ordering. Materialize the data with:

make -C etl BASE=https://linkeddata.lt/

🤖 Generated with Claude Code