Investigate: Adopt Felles datakatalog classification (DCAT-AP-NO, EU Data Theme, LOS)

IMPLEMENTATION RULES: Before implementing this plan, read and follow:

• WORKFLOW.md - The implementation process
• PLANS.md - Plan structure and best practices

Status: Backlog

Goal: Decide how Atlas should align its source-classification scheme (PLAN-007 manifest.yml tags) with the controlled vocabularies used by Felles datakatalog (DCAT-AP-NO), so Atlas's catalogue is interoperable with national and EU dataset discovery without giving up the domain-precise vocabulary the contributor frontend benefits from.

Last Updated: 2026-05-01

Origin: PLAN-007 (Data display — open by default) introduces manifest.yml per source with four tag namespaces (provider, topic, geo, cadence). The user asked how Atlas should learn from data.norge.no's classification work — specifically Utgiver, Tema, EU-tema, and modelldcat-ap-no information models. This investigation evaluates each system, recommends an adoption path, and sequences the work behind PLAN-007 so it doesn't block delivery of the open-by-default frontend.

---

What Felles datakatalog actually uses

Three distinct classification systems are surfaced as filter facets on data.norge.no, plus the metadata standard underneath:

Their facet	What it is	Vocabulary	URI prefix
Tema	Norwegian topic taxonomy for public services	LOS — SKOS vocabulary maintained by Digdir	https://psi.norge.no/los/
EU-tema	EU's 13-bucket data-theme vocabulary	DCAT-AP dcat:theme (EU Publications Office)	http://publications.europa.eu/resource/authority/data-theme/{CODE}
Utgiver	Publisher organisation	DCAT-AP dct:publisher	Brreg orgnr
(the catalogue itself)	DCAT-AP-NO 2.0 — Norwegian profile of EU's DCAT-AP	RDF/Turtle/JSON-LD	n/a

Separately, modelldcat-ap-no is the spec for information models (schemas, not datasets) — the URL the user linked points to one of these. It's an adjacent but distinct concern from dataset classification.

The 13 EU Data Theme codes:

Code	Label (EN)	Norwegian gloss
AGRI	Agriculture, fisheries, forestry and food	Landbruk, fiske, skogbruk og mat
ECON	Economy and finance	Økonomi og finans
EDUC	Education, culture and sport	Utdanning, kultur og sport
ENER	Energy	Energi
ENVI	Environment	Miljø
GOVE	Government and public sector	Forvaltning og offentlig sektor
HEAL	Health	Helse
INTR	International issues	Internasjonale spørsmål
JUST	Justice, legal system and public safety	Justis, rettssystem og offentlig sikkerhet
REGI	Regions and cities	Regioner og byer
SOCI	Population and society	Befolkning og samfunn
TECH	Science and technology	Vitenskap og teknologi
TRAN	Transport	Transport

---

Questions to Answer

1. Which Atlas topics map to which EU Data Theme? How many Atlas topics fall outside the 13 buckets cleanly, and what do we do about them?
2. Is LOS a useful classification for Atlas datasets, or only for service consumers? LOS is a public-services taxonomy organised around life events (å få barn, flytte, etc.), not raw statistics — but Samfunnspuls services that consume Atlas data could be LOS-classified.
3. Should Atlas publish its own DCAT-AP-NO catalogue? This would make Atlas harvestable by data.norge.no and (transitively) by data.europa.eu. What's the minimum surface area — a single /catalog.ttl endpoint? A mart_meta_dcat model? When does this become worth doing?
4. Should the contributor and customer frontends expose EU theme as a filter alongside Atlas's domain topic, or instead of it? EU theme is coarser; UX may benefit from the finer-grained Atlas vocabulary; both is possible but adds chrome.
5. Where does the mapping live? Per-source manifest.yml field (eu_theme: SOCI)? Lookup seed (eu_data_theme.csv) joined into mart_meta_sources? Both?
6. Where do modelldcat-ap-no information models fit, if at all? Atlas's api_v1.* endpoints are in some sense information models — could mart_meta_endpoints eventually emit modelldcat-ap-no descriptions? Probably out of scope short-term but worth noting.

---

Current state

After PLAN-007 Phase 2 lands, Atlas has:

• 21 manifest.yml files under atlas-data/ingest/src/sources/ with 4 tag namespaces (provider, topic, geo, cadence).
• Atlas-domain topic vocabulary: ngo-supply, reference, income, education, health, social, demographics.
• Publisher field already keyed to Norwegian organisation names (SSB, FHI, Norges Røde Kors).
• No EU Data Theme tags. No LOS tags. No DCAT-AP-NO catalogue endpoint.

---

Mapping Atlas topic → EU Data Theme (initial pass)

Speculative, to anchor discussion:

Atlas tags.topic	EU Data Theme	Notes
education	EDUC	Clean fit.
health	HEAL	Clean fit.
demographics	SOCI	EU Data Theme SOCI is "Population and society" — explicit fit.
social	SOCI	Same bucket as demographics — collapses two Atlas topics.
income	SOCI or ECON	Lavinntekt is socioeconomic; ECON is "Economy and finance" (more macro). SOCI is the better fit because Atlas's income data is household-level inequality, not GDP/finance.
reference	(no fit)	Reference data (kommune codes, classifications) is structural, not domain. Could go to GOVE if forced.
ngo-supply	(no fit)	NGO/civil-society data has no dedicated EU theme. SOCI is closest; GOVE if framed as public-sector adjacent.

Conclusion: 5 of 7 Atlas topics map cleanly; 2 don't. The collapse (income+social+demographics → SOCI) is information loss for Atlas's UX, but acceptable for federated discovery where coarser is fine.

---

Recommendation (provisional — to be confirmed in this investigation)

Add eu_theme as a parallel namespace in manifest.yml. Keep topic as Atlas-domain.

Rationale:

• EU Data Theme is the lingua franca for federated dataset discovery.
• Cost is ~1 line per manifest + a single seed CSV with 13 rows.
• Preserves Atlas's finer-grained topic for the /data UX (filter by income vs. social vs. demographics is more useful to a user than collapsing all three to SOCI).
• Unlocks two things later:
  1. A future /catalog.ttl (or mart_meta_dcat) endpoint that data.norge.no can harvest.
  2. Link-out from Atlas's per-source page to the same dataset's neighbours on data.europa.eu.

Don't adopt LOS for dataset classification. LOS is the wrong abstraction layer — it classifies services and life events, not raw data. If Atlas later describes Samfunnspuls services or other downstream consumers, LOS is the right vocabulary for those entities, not for the underlying datasets.

Defer DCAT-AP-NO publishing as a separate workstream. Material payoff (national catalogue inclusion → discoverability), but it's a meaningful chunk of work (RDF emission, harvest URL, conformance testing). Should land as its own PLAN after PLAN-007 ships.

Defer modelldcat-ap-no entirely unless and until Atlas has a use case for publishing schemas as RDF. The standard is real, but Atlas's api_v1.* is already self-describing via PostgREST/OpenAPI; the duplication isn't justified yet.

---

Concrete shape if we go ahead

manifest.yml schema amendment (Phase 2 of PLAN-007 or a follow-up phase):

source_id: ssb-08764
# ... existing fields ...
tags:
  provider: ssb
  topic: income            # Atlas-domain, unchanged
  geo: kommune
  cadence: annual
eu_theme: SOCI             # NEW — DCAT-AP data-theme code, single value
los_theme: null            # NEW — usually null; populated only where a clean LOS concept exists

Plus a seed at atlas-data/dbt/seeds/eu_data_theme.csv:

code,uri,label_en,label_no
AGRI,http://publications.europa.eu/resource/authority/data-theme/AGRI,Agriculture fisheries forestry and food,Landbruk fiske skogbruk og mat
…
SOCI,http://publications.europa.eu/resource/authority/data-theme/SOCI,Population and society,Befolkning og samfunn
…

Joined into mart_meta_sources so the customer frontend can render both Atlas-topic and EU-theme filters.

---

Sequencing

This investigation produces a one-page outcome note + an amendment to PLAN-007 (add eu_theme field, add eu_data_theme.csv seed). It does NOT block PLAN-007 Phase 2.5+ — those phases land first, this amendment lands as a Phase 2.7 follow-up or a small standalone PLAN.

The DCAT-AP-NO publishing question becomes its own INVESTIGATE-dcat-ap-no-publishing.md once Atlas has at least one external user asking for harvestability.

---

Open questions for the user

1. EU theme cardinality: single value per source (simplest), or array (more accurate when a source genuinely spans two themes — e.g. ssb-09429 educational attainment by kommune touches EDUC + SOCI)? Recommendation: single value, primary theme only, until proven insufficient.
2. LOS confirmation: agree that LOS doesn't fit Atlas dataset classification? (Or is there a use case I'm missing — e.g. "this dataset is relevant to the Få barn life event"?)
3. DCAT-AP-NO publishing timing: confirm this is a separate later PLAN, not PLAN-007 scope.

---

Sources

• Felles datakatalog Search API documentation
• DCAT-AP-NO 2.0 specification (Informasjonsforvaltning)
• EU Publications Office Data Theme vocabulary
• LOS vocabulary documentation (Digdir)
• modelldcat-ap-no specification
• DCAT-AP 3.0 (SEMIC EU)