MAIA-Madrid-IA/cibelex-graph-core-sampler

Cibelex Knowledge Graph — Core Sampler (v1.0.3)

A curated subset of the Cibelex regulatory database of the Municipality of Madrid, annotated with the LoRO ontology and serialised as named-graph Turtle files. Built and maintained by the MAIA initiative (Ayuntamiento de Madrid).

Why "core sampler"? The release covers a representative slice of the Cibelex database — municipal ordinances, regulations and organic regulations — deliberately including multiple historical versions of the same legal texts to support temporal reasoning. It is intentionally smaller than the full Cibelex catalog, in the spirit of a geological core sampler that captures every layer of interest in a manageable cross-section.

At a glance

• Total triples: 350,258
• Named-graph files: 9
• Bundled ontology: LoRO v1.0 (ontology/loro.ttl) — snapshot of Zenodo deposit
• License: Creative Commons Attribution 4.0 International (CC-BY 4.0)
• Release date: 2026-05-27
• Canonical URI namespace: http://maia.madrid.es/ontologies/def/local-regulations#

Files

The repository is laid out so each consumer finds their format in a stable place:

.
├── README.md                                 ← this card
├── datapackage.json                          ← Frictionless Data descriptor
├── MANIFEST.json                             ← per-file inventory + counts
├── ttl/                                      ← canonical RDF (rdflib / SPARQL)
│   ├── event.ttl
│   ├── feature.ttl
│   ├── legal-resource.ttl
│   ├── legal-resource-subdivision.ttl
│   ├── legal-value.ttl
│   ├── organization.ttl
│   ├── person.ttl
│   ├── role.ttl
│   └── taxonomy.ttl
├── ontology/loro.ttl                         ← bundled LoRO v1.0
└── data/                                     ← Parquet (load_dataset)
    └── train-00000-of-00001.parquet

Per-graph TTL

File	Triples	Size
ttl/event.ttl	324	0.1 MB
ttl/feature.ttl	121,020	7.7 MB
ttl/legal-resource.ttl	21,556	6.0 MB
ttl/legal-resource-subdivision.ttl	185,113	55.1 MB
ttl/legal-value.ttl	5	0.0 MB
ttl/organization.ttl	2,088	0.5 MB
ttl/person.ttl	121	0.0 MB
ttl/role.ttl	360	0.1 MB
ttl/taxonomy.ttl	19,671	3.1 MB
ontology/loro.ttl	ontology axioms	bundled snapshot

Quick start

The simplest way to load this graph is the companion container loro-fuseki, which downloads the dataset, mounts the TTL files, builds a Lucene full-text index and exposes a SPARQL endpoint:

git clone https://github.com/AyuntamientoMadrid/loro-fuseki
cd loro-fuseki
./scripts/run-local.sh           # downloads this dataset + builds + starts
# → SPARQL endpoint at http://localhost:3030/loro/sparql

For an LLM-friendly access layer (Claude Code, Cursor, Claude Desktop), the cibelex-mcp-fuseki MCP server wraps 21 high-level retrieval tools on top of that Fuseki endpoint.

You can also load the graph directly with rdflib:

from huggingface_hub import snapshot_download
from rdflib import Graph

local = snapshot_download(repo_id="MAIA-Madrid-IA/cibelex-graph-core-sampler", repo_type="dataset")

g = Graph()
for ttl in [
    "event.ttl", "feature.ttl", "legal-resource.ttl",
    "legal-resource-subdivision.ttl", "legal-value.ttl",
    "organization.ttl", "person.ttl", "role.ttl", "taxonomy.ttl",
]:
    g.parse(f"{local}/ttl/{ttl}", format="turtle")

print(f"Loaded {len(g):,} triples")

Loading via datasets

A unified triple table is also published as a single Parquet shard at data/train-00000-of-00001.parquet (350,258 rows, 8.7 MB). It contains the same triples as the TTL files in lossless form (one row per triple, with the named-graph IRI in the graph column), so you can round-trip back to RDF.

from datasets import load_dataset

ds = load_dataset("MAIA-Madrid-IA/cibelex-graph-core-sampler", "triples", split="train")
print(ds)                  # 7 columns: subject, predicate, object, object_type,
                           # object_lang, object_datatype, graph
print(ds[0])

Or directly with polars / pandas:

import polars as pl
df = pl.read_parquet("hf://datasets/MAIA-Madrid-IA/cibelex-graph-core-sampler/data/train-00000-of-00001.parquet")

Data scope and provenance

• Source corpus: the Cibelex regulatory database of the Municipality of Madrid — the official catalog of regulatory documents enacted by the City Council and its delegated bodies.
• Annotation: the LoRO ontology extends the European Legislation Identifier (ELI) to cover document types specific to local government. This release is annotated with LoRO v1.0.
• Coverage: a representative subset focused on municipal ordinances, regulations and organic regulations. Multiple historical versions of selected texts are intentionally retained to enable temporal queries (e.g., what an article said before and after an amendment).
• Future versions will progressively extend coverage toward the full Cibelex catalog.

Persons in the dataset

The person.ttl graph uses role-based descriptors for public officials (for example, Concejal del Ayuntamiento de Madrid núm. 015 (PP)) rather than civil names. The reproducible build writes the audited mapping locally under reports/anonymization_mapping.json, but that report is not part of the published dataset. No private IDs (DNI/NIE), bank accounts (IBAN) or sensitive personal attributes are included.

Feature descriptions held back pending legal review (v1.0.2)

dcterms:description literals on geonames:Feature resources are not included in this release. They contained encyclopedic text about municipal places (libraries, parks, cultural centres, etc.) that surfaced names of public officials, named-after toponyms and other persons in narrative form. Until the data protection assessment validates a clear policy for that field, the conservative compliance posture for this release is to remove the field entirely from feature.ttl. Schema-relevant predicates on each geonames:Feature remain published unchanged: geonames:name, geonames:alternateName, wgs84_pos:lat / wgs84_pos:long (typed xsd:float), geonames:featureClass, dcterms:identifier, and the links to other named graphs. Once descriptions are reviewed and reincorporated they will be added back in a later minor release.

Coordinate normalisation

Feature coordinates are normalised during the reproducible build. Legacy raw predicates geonames:latitude and geonames:longitud are rewritten to WGS84 wgs84_pos:lat and wgs84_pos:long; comma-decimal string literals such as "40,384304" are converted to dot-decimal xsd:float literals. The local build report is written to reports/geo_normalization_report.json.

Named graphs

Each TTL file corresponds to one named graph in the dataset:

File	Named graph URI
ttl/event.ttl	http://maia.madrid.es/ontologies/loro/event
ttl/feature.ttl	http://maia.madrid.es/ontologies/loro/feature
ttl/legal-resource.ttl	http://maia.madrid.es/ontologies/loro/legal-resource
ttl/legal-resource-subdivision.ttl	http://maia.madrid.es/ontologies/loro/legal-resource-subdivision
ttl/legal-value.ttl	http://maia.madrid.es/ontologies/loro/legal-value
ttl/organization.ttl	http://maia.madrid.es/ontologies/loro/organization
ttl/person.ttl	http://maia.madrid.es/ontologies/loro/person
ttl/role.ttl	http://maia.madrid.es/ontologies/loro/role
ttl/taxonomy.ttl	http://maia.madrid.es/ontologies/loro/taxonomy
ontology/loro.ttl	http://maia.madrid.es/ontologies/loro/ontology

Citation

When using this dataset in research, please cite both the LoRO ontology and the dataset:

@misc{loro_ontology_2026,
  author    = {{Ayuntamiento de Madrid}},
  title     = {{LoRO --- Local Regulations Ontology}},
  year      = {2026},
  publisher = {Ayuntamiento de Madrid},
  doi       = {10.5281/zenodo.20076577}
}

@misc{cibelex_graph_core_sampler_2026,
  author    = {{Ayuntamiento de Madrid}},
  title     = {{Cibelex Knowledge Graph --- Core Sampler}},
  year      = {2026},
  publisher = {Ayuntamiento de Madrid},
  url       = {https://huggingface.co/datasets/MAIA-Madrid-IA/cibelex-graph-core-sampler}
}

License

Released under Creative Commons Attribution 4.0 International (CC-BY 4.0). You are free to share and adapt the data, including for commercial use, provided you give appropriate credit and indicate any changes you make.

Maintenance

Maintained by the MAIA team (Madrid Artificial Intelligence) at the IT Department (IAM) of the Municipality of Madrid. Issues, suggestions and contributions are welcome via the project landing page.

Disclaimer

This repository provides semantic resources intended for regulatory analysis. Its intended use is limited to exploratory analysis, research, and prototyping purposes.

It does not constitute an artificial intelligence system for automated administrative decision-making, the assessment of citizens' rights, or legal advice.

Any use of these resources within automated systems, including those integrating artificial intelligence models, is the sole responsibility of the final integrator or developer, who must ensure compliance with all applicable regulations.

Aviso legal

Este repositorio proporciona recursos semánticos destinados al análisis de normativa. Su uso previsto se limita a fines de análisis exploratorio, investigación y prototipado.

No constituye un sistema de inteligencia artificial para la toma de decisiones administrativas automatizadas, la evaluación de derechos de los ciudadanos ni el asesoramiento jurídico.

Cualquier uso de estos recursos en sistemas automatizados, incluidos aquellos que integren modelos de inteligencia artificial, es responsabilidad exclusiva del integrador o desarrollador final, quien deberá garantizar el cumplimiento de la normativa aplicable.