generated: '2026-08-09'
method: derived
source: openapi/i6eal-open-ai-data-api-openapi.json
also_derived_from:
  - json-ld/i6eal-open-ai-data-api-dcat.jsonld
  - examples/i6eal-open-ai-data-api-supervision-methodology.json
summary: >-
  The model is a two-level publishing graph, not a transactional object graph. A single
  Catalogue holds 16 Datasets; each Dataset holds one or more Distributions (a JSON index,
  zero-or-more CSV tables, an optional Atom change feed, a methodology document). There are
  no ids to POST, no foreign keys between API resources, and no query parameters — the
  "primary key" of the whole API is the distribution URL, and the DCAT catalogue is the
  authoritative map of those URLs.
entities:
  - name: Catalogue
    type: dcat:Catalog
    schema: DcatCatalogue
    operation: getOpenDataCatalogue
    url: https://i6eal.de/data/catalog/dcat.jsonld
    identifier: '@id'
    description: >-
      The DCAT JSON-LD catalogue. Root discovery document; carries dct:title,
      dct:description, dct:publisher (i6eal, foaf:Organization), dct:modified, dct:rights
      (de + en) and i6eal:compilationLicense.
    count: 1
  - name: Dataset
    type: dcat:Dataset
    identifier: 'dct:identifier'
    count: 16
    description: >-
      One tracked evidence collection (procurement chronicle, budget monitor, case-law
      monitor, …). Every dataset carries bilingual dct:title/dct:description, dct:modified,
      dcat:landingPage, dct:source (the official upstream hosts), dct:rights, dct:provenance,
      schema:creditText and the i6eal-private observedSince / collectionState /
      collectionCadence / sourceRightsStatus. 12 of 16 also carry dct:temporal +
      schema:temporalCoverage; the four without it state that source coverage varies.
    key_fields:
      - dct:identifier
      - i6eal:observedSince
      - i6eal:collectionState
      - i6eal:collectionCadence
      - dct:temporal
  - name: Distribution
    type: dcat:Distribution
    identifier: dcat:accessURL
    count: 82
    description: >-
      A concrete downloadable file. Each carries dct:title (de + en), dct:format,
      dcat:mediaType, dcat:accessURL and dcat:downloadURL — and every one of them is also an
      OpenAPI operation, which is why the catalogue and the OpenAPI describe the same 82/83
      URL set from two directions.
  - name: DatasetIndex
    schema: DatasetIndex
    media_type: application/json
    description: >-
      The full JSON projection of a dataset. Shared envelope: schemaVersion (payload
      contract, integer 1 observed on every dataset), generatedAt (build timestamp),
      trackingSince (observation start), methodology (how it is collected and what it does
      not claim), downloads (named sibling distribution paths). All remaining top-level keys
      are dataset-specific — observed examples include counts, sourcePolicy (officialOnly +
      allowedHosts), sourceCollection, collectionStatus, sourceHealth, graph, events,
      dossierSlugs, materialFingerprint and sourceFingerprint.
    operations: 16
  - name: CsvTable
    schema: CsvTable
    media_type: text/csv
    description: >-
      RFC 4180, header row, UTF-8. The relational face of a dataset — the flat tables an
      analyst joins. Table names are dataset-specific (events, changes, titles, series,
      revisions, systems, links, entities, edges, decisions, passages, authorities,
      instruments, relationships, work-items, providers, evidence, countries, observations,
      questions, answers, platforms, flows, timeline, devices, actions, studies).
    operations: 37
  - name: AtomFeed
    schema: AtomFeed
    media_type: application/atom+xml
    description: >-
      Atom 1.0 feed of observed material changes to a dataset. Published by 11 of the 16
      datasets, under two path spellings — atom.xml (5) and feed.atom (6). Entry ids follow
      the pattern urn:i6eal:<dataset>:event:<date>:<kind>-<hash>.
    operations: 11
  - name: Methodology
    media_type: application/json
    description: >-
      Per-dataset statement of how the data is collected and what it does and does not
      claim. Published by all 16 datasets as methodology.json.
    operations: 16
  - name: ArchiveDocument
    media_type: application/json
    description: >-
      Rolling archive of superseded observations, published by one dataset only
      (ki-auftragsaenderungen via getProcurementObservedChangesArchive).
    operations: 1
  - name: DatasetSchema
    media_type: application/json
    description: >-
      Explicit payload schema, published by one dataset only (ki-laender-lagebild via
      getGermanStateAiEvidenceLayersSchema). The other 15 datasets describe their shape in
      prose inside methodology.json plus the DCAT entry.
    operations: 1
relationships:
  - from: Catalogue
    to: Dataset
    kind: has_many
    via: dcat:dataset
    cardinality: 1..16
  - from: Dataset
    to: Distribution
    kind: has_many
    via: dcat:distribution
    cardinality: 1..N
  - from: Dataset
    to: DatasetIndex
    kind: has_one
    via: index.json distribution
  - from: Dataset
    to: CsvTable
    kind: has_many
    via: '*.csv distributions'
  - from: Dataset
    to: AtomFeed
    kind: has_one
    via: atom.xml | feed.atom
    optional: true
    note: present on 11 of 16 datasets
  - from: Dataset
    to: Methodology
    kind: has_one
    via: methodology.json
  - from: DatasetIndex
    to: Distribution
    kind: has_many
    via: downloads
    note: the index names its own sibling distributions, so one fetch reveals the rest
  - from: Dataset
    to: Publisher
    kind: belongs_to
    via: 'dct:publisher -> https://i6eal.de/#organization'
  - from: Distribution
    to: Operation
    kind: has_one
    via: dcat:accessURL == OpenAPI path
    note: >-
      The join between the two machine-readable surfaces. Resolving a DCAT
      dcat:accessURL against the OpenAPI paths object yields the operationId for that file.
identifier_conventions:
  dataset_id: >-
    Lowercase hyphenated English slug in DCAT (procurement-chronicle, …) but a German slug in
    the URL path (/data/ki-auftragschronik/) and a third form in the payload datasetId
    (ki-aufsichtsmonitor-de-eu). All three are stable; do not assume they match.
  atom_entry_id: 'urn:i6eal:<dataset>:event:<YYYY-MM-DD>:<kind>-<hash16>'
  integrity: materialFingerprint and sourceFingerprint are SHA-256 hex digests per index
  joining_rule: >-
    From llms.txt — records are joined on exact official identifiers only (TED/OCDS notice
    ids, MaKI record ids, CORDIS PIC, ECLI, federal budget title numbers). A shared or
    similar name never creates a link between two records.
cross_links:
  conventions: conventions/i6eal-open-ai-data-api-conventions.yml
  json_ld: json-ld/i6eal-open-ai-data-api-dcat.jsonld
  examples: examples/i6eal-open-ai-data-api-examples.yml