The Egyptian Legal Citation Graph
We scanned the Egyptian legal corpus for citation mentions, resolved what we could verify, and threw out what we couldn't.
- 1Of 1.5 million citation-shaped mentions found in the text, only 17.4% (261,770) resolve to exactly one document already in the corpus and become an edge. 17.4% collide with more than one candidate sharing the same number and year and are dropped rather than guessed; 59.3% cite something outside this corpus entirely, most likely older instruments not yet ingested; and the remaining 5.9% are duplicate citations to a target the same document already cited elsewhere, collapsed into the existing edge instead of counted again.
- 2Citation activity concentrates hard: the 12 most-cited instruments account for 43.3% of all resolved edges, and the top 100 account for 68.3%, even though 6,768 distinct statutes receive at least one citation somewhere in the graph.
- 3The most-cited instruments are not sweeping codifications. They're narrow administrative and regulatory decrees, things like company branch licensing or land registration procedure, cited over and over by structurally similar rulings. That tracks with the corpus itself: 57% of Egypt's indexed documents carry an administrative legal_domain tag.
- 499.9% of resolved edges point to legislation, not to other court rulings, even though 57.7% of the documents doing the citing are themselves judgements. Courts cite statutes far more than they cite each other, at least among what resolves inside this corpus.
1. Building a real citation graph instead of assuming one
Dike's corpus stores an embedded graph field on every Egyptian document (cites, cited_by, amends, amended_by), but until now it sat empty across the corpus. The citation parser that powers /v1/cite already knew how to recognize Egyptian citation formats (the "Law No. N of year Y" pattern, cassation appeal numbers, ministerial and presidential decrees, fatawa numbers), but only for parsing a single string typed by a caller. We extended it to scan full document text for every citation-shaped mention, then tried to resolve each one against an index built from the corpus itself: same document type bucket, same year, same number. A mention that matches exactly one candidate becomes an edge. A mention that matches more than one is left out. We'd rather publish a smaller graph than a wrong one.
1,504,616 mentions extracted → 261,770 resolved (17.4%), 261,720 ambiguous and dropped (17.4%), 892,887 unresolved (59.3%), 87,775 duplicate mentions collapsed into an existing edge (5.8%), 464 self-references dropped (0.03%). Every mention is accounted for in exactly one bucket; the unresolved majority is the honest ceiling on what a citation graph can cover until older source material gets ingested.
2. Where citation activity concentrates
The table below shows the most-cited instruments by resolved in-degree. We're identifying them by their formal citation (number and year) rather than a popular name, because this corpus's title field holds a generated summary of each document's content, not an official short title, and issuing_body is empty for most legislation records. What we can say with confidence is how often each one is cited and by how many distinct documents.
| Citation | Year | Cited By | Share of All Resolved Edges |
|---|---|---|---|
| قانون/قرار رقم 57 لسنة 1959 | 1959 | 35,337 documents | 13.5% |
| قانون/قرار رقم 43 لسنة 1979 | 1979 | 26,106 documents | 10.0% |
| قانون/قرار رقم 47 لسنة 1972 | 1972 | 12,552 documents | 4.8% |
| قانون/قرار رقم 23 لسنة 1992 | 1992 | 6,650 documents | 2.5% |
| قانون/قرار رقم 46 لسنة 1972 | 1972 | 5,230 documents | 2.0% |
| قانون/قرار رقم 53 لسنة 1966 | 1966 | 5,189 documents | 2.0% |
| قانون/قرار رقم 116 لسنة 1983 | 1983 | 5,050 documents | 1.9% |
3. An administrative-law skew, not a codification story
A citation graph built from a corpus that's 57% administrative rulings produces an administrative-law graph. The top anchor, cited by over 35,000 documents, isn't a landmark code, it's a narrow 1959 decree that keeps surfacing because large numbers of structurally similar administrative rulings invoke the same procedural basis. That's not a flaw in the extraction, it's what happens when you measure citation behavior honestly instead of assuming courts mostly argue over civil or commercial doctrine. In this corpus, most of them don't.
4. What this means for retrieval and agents
For anyone building retrieval over this corpus, the concentration finding is directly useful: a small set of anchor instruments (12 of them cover 43.3% of resolved citations) is disproportionately likely to be relevant context for administrative queries, and dense retrieval that treats the whole corpus as uniformly likely to matter will underweight them. The honest caveat matters just as much: this graph currently only covers Egypt (Saudi's citation conventions aren't supported by the parser yet), and an edge's absence doesn't mean a citation doesn't exist in the real world, only that we couldn't verify it against this corpus without guessing.
5. Limitations and what's next
Three gaps are worth naming plainly. First, 59.3% of extracted mentions don't resolve, most likely because they cite statutes or rulings from before this corpus's ingestion window, though some are almost certainly extraction misses worth revisiting. Second, ambiguity resolution is currently binary: a mention with two or more candidates is dropped entirely rather than resolved using surrounding context, which is conservative but throws away 261,720 potentially real edges. Third, Saudi Arabia isn't represented in this graph at all; its citation formats (royal decree numbering, for instance) aren't yet covered by the extraction patterns. All three are addressable in a follow-up pass, and we'd rather report the graph as it stands today than wait for a complete one.
Bibliographic Reference (BibTeX)
@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}
}