Datasets:

SNTIQ-Team
/

amtsgraph

license: other
license_name: sntiq-dual-by-audience
license_link: https://github.com/SNTIQ-Team/amtsgraph/blob/main/LICENSING.md
language:
  - de
pretty_name: Amtsgraph — German authority and EU institution graph
tags:
  - germany
  - european-union
  - eu-institutions
  - legal
  - government
  - courts
  - public-administration
  - graph
size_categories:
  - 100K<n<1M
configs:
  - config_name: authorities
    data_files: authorities.jsonl
  - config_name: court_chains
    data_files: court_chains.jsonl
  - config_name: competences
    data_files: competences.jsonl
  - config_name: edges
    data_files: edges.jsonl
  - config_name: places_courts
    data_files: places_courts.jsonl
  - config_name: municipalities
    data_files: municipalities.jsonl
  - config_name: matters
    data_files: matters.jsonl
  - config_name: caveats
    data_files: caveats.jsonl

Amtsgraph

The open, validated graph of German public authorities and a curated, source-backed EU institutional layer — who is competent, where, and for what, without flattening distinct legal relations into a fake hierarchy.

Built 2026-07-15 from official registers (Orts- und Gerichtsverzeichnis, PVOG/FITKO, Bundesagentur für Arbeit, Destatis, BayernPortal, OpenPLZ, xRepository), plus EU primary law and official institution pages for the reviewed 11-node EU overlay. Every build passes a validation gate (coverage, collision, contiguity, benchmark and EU cross-hierarchy checks) before release; every record carries provenance (source, URL, fetch timestamp).

Files

File	Rows	Content
`authorities.jsonl`	19,757	all authorities with contacts, provenance, external IDs
`court_chains.jsonl`	338,873	instance chains per (plz, ortk, matter)
`competences.jsonl`	77,125	kind × area assignments (rank 0 = apply here)
`edges.jsonl`	8,424	typed, weighted relations with provenance and legal limits
`places_courts.jsonl`	14,057	official court-resolution place register
`municipalities.jsonl`	10,950	AGS/ARS spine with postal codes (M:N)
`matters.jsonl`	36	legal-matter taxonomy (14 core + special)
`caveats.jsonl`	5,385	warnings that must be surfaced with answers

Resolution logic

Courts: look up places_courts by postal code (+ locality when one code spans several court districts), then court_chains by (plz, ortk, matter) — rows ordered by position are the full instance chain (role court) plus prosecution offices (role prosecution). Join authorities for contacts and filing addresses.
Agencies: map the user's municipality to its ags via municipalities, then filter competences by kind and area (gemeinde = 8-digit AGS, kreis = 5-digit prefix, land = 2-digit). rank 0 = the office to apply at; rank 1 = supervisory body.
Always join caveats for the scope you resolved — they flag genuine ambiguities and register gaps. For legal use, an honest warning beats a confident wrong answer.
The organisational web (Bavaria): traverse edges with relation='parent' for department hierarchies, 'supervision' for oversight, 'appeal' for derived court appeal routes.
The curated EU overlay uses eu_institution, eu_body and eu_court nodes. Its edges are deliberately limited to the relation named in each row (institutional_part, political_accountability, judicial_review, co_legislation, reporting_accountability, financial_audit, maladministration_review, sectoral_oversight, or a judicial appeal). Read note and source_url: there are no EU-to-German supervisory edges. Treaty-defined preliminary rulings can bind the referring national court on the EU-law question, but are judicial cooperation rather than an appeal from its judgment; that class-level procedure is documented, not guessed as a blanket authority hierarchy. Each of the 11 EU nodes has an officially sourced postal address and telephone. E-mail and fax remain null when the institution publishes only a contact form; the Court of Justice and General Court use their distinct official registry mailboxes, and CURIA states that fax is no longer used.

Null semantics

Empty fields are deliberate, not dirty data:

legal_form is only defined for Jobcenter (gE = joint federal-municipal institution, zkT = municipal provider); for every other kind it is null.
For courts, the full visiting address (street + postal code + city) lives in street as published by the justice register, and the separate postal-box address in postal_address; the plz/city columns are filled where sources provide them as discrete fields (agencies), null otherwise.
valid_to is null for active records; a timestamp marks retired ones (kept for traceability — e.g. offices replaced by a more precise department).
matter on edges is null for non-court relations and the EU judicial overlay; EU access and jurisdiction remain procedure-specific.
external_ids may be empty for organisational units that have no official register identity of their own.

Important notes

General information, not legal advice; statutory exceptions to competence rules exist that no dataset can capture.
Department-level depth covers Bavaria's 96 Kreisverwaltungsbehörden (pilot); other Länder resolve at authority level.
Underlying records originate from official public registers; their respective terms govern redistribution (see the source registry in the GitHub repository). Dual-licensed by audience: non-commercial / human-rights / research use under Apache-2.0 (code) + CC BY-NC-SA 4.0 (data); commercial, corporate and governmental use under the SNTIQ-CM License v1.0 (free written permission, non-harm covenant, attribution, reciprocity) — see LICENSING.md in the GitHub repository.

Citation

@misc{amtsgraph2026,
  title   = {Amtsgraph: an open validated graph of German public
             authorities, court competences and organisational structure},
  author  = {Glushkov, David and {SNTIQ}},
  year    = {2026},
  url     = {https://github.com/SNTIQ-Team/amtsgraph}
}

Source code, pipeline and full documentation: https://github.com/SNTIQ-Team/amtsgraph