dikelab/eg-legal-citation-graph

Egyptian Legal Citation Graph

103,711 citation edges pulled from Egypt's legal corpus at Dike: legislation citing legislation only - statutes, decrees, and regulations referencing one another (e.g. an amending law citing the law it amends). Companion dataset to this report.

This release is scoped to legislation-legislation edges. Judgements, fatawa, and Economic Court rulings (and any edge touching them) are not included here.

Files

graph.jsonl (86,589 rows) - one row per legislation document, with its citation edges embedded.

{"canonical_id":"eg-legislation-100087","doc_type":"legislation","year":1978,"citation":"قانون/قرار رقم 10 لسنة 1978","status":"in_force","cites":["eg-legislation-106462"],"cited_by":[]}

field	what it is
canonical_id	stable ID for this document
doc_type	always legislation in this release
year	citation year, when known - missing (not null) on about 21% of rows
citation	a display label, not a parseable field - formatting varies, some rows have a date segment and some don't. Use year for anything programmatic
status	always in_force in this release, see below
cites	canonical_ids this document cites (edges out)
cited_by	canonical_ids that cite this document (edges in)

103,711 citation edges total, counted across all cites arrays. If a document cites the same target twice, that's one entry, not two - edge count means "cites at least once," not raw mention count. Every ID in cites/cited_by is itself a canonical_id present as a row in this file.

How the IDs and types map

This release carries one canonical_id prefix, tied to one doc_type and one fixed status:

prefix	doc_type	status	what it is
eg-legislation-	legislation	in_force	statutes, decrees, regulations

status doesn't vary independently, it's just doc_type in disguise. Don't filter on it expecting more granularity than that today.

What's not here

No document text. This dataset is graph structure and citation labels only, nothing else:

• no full_text, verbatim_text, or articles
• no summary, keywords, topics, parties (all derived NLP output)
• no title field - in this corpus it's an LLM-generated summary, not a real document name, so we're not passing it off as one
• no internal storage paths or pipeline metadata

If you need the actual text, search, or grounded Q&A over this corpus, that's what the Dike API is for.

License

CC-BY-4.0.

Citation

@article{dike2026egyptcitationgraph,
  title={The Egyptian Legal Citation Graph},
  author={Dike Research Team},
  year={2026},
  publisher={Dike AI},
  url={https://dike.it.com/research/egyptian-legal-citation-graph-2026}
}