VaidhyaMegha/legal-judgments-kg

Indian Supreme Court Judgments Knowledge Graph

4,462 nodes. 8,363 edges. 589 Supreme Court of India judgments from 2016 — cases, the judges who decided them, the parties before them, the acts they cite down to the section, and the legal topics they turn on.

Built with Samyama Graph. Loader and ETL: samyama-ai/legal-judgments-graph-kg.

Attribution — required

This dataset is CC-BY-4.0, so attribution is a condition of use, not a courtesy:

Indian Supreme Court Judgments, curated by Dattam Labs, accessed from the AWS Open Data Registry. Licensed CC-BY-4.0.

That requirement travels with the data — anything you redistribute or build on must carry it.

Data	CC-BY-4.0 — attribution required, redistribution permitted
Original source	AWS Open Data Registry, curated by Dattam Labs
Intermediate	Shreyasrao/Indian-law-supreme-court-judgements-2016 (rev e928c72019d6)
Loader code	Apache-2.0, in the GitHub repo. This does not govern the data.

Licence checked at the source, not inferred: the AWS Open Data Registry entry states CC-BY-4.0 and explicitly permits redistribution and derivative works with attribution.

What this is

Every judgment the Supreme Court of India handed down in 2016 that the source corpus covers, as a property graph rather than a pile of documents.

The useful part is that the cited section lives on the edge, not on the case. So "how many judgments cite IPC §302?" is a single hop, and questions that span cases — which judges sit together, which statutes are cited side by side, which laws reach the widest range of subject matter — are traversals rather than joins.

Files

Nodes carry an id; edges reference those ids as src and tgt. Join on id.

nodes/

File	Rows	Columns
topic.csv	2,291	id, category, text
party.csv	1,102	id, name
case.csv	589	id, case_id, month, title, year
act.csv	446	id, name
judge.csv	34	id, name

case_id is the source judgment filename (e.g. 2016-1-1-17-en.md). It is named case_id rather than id because the internal node id already occupies that column.

edges/

File	Rows	Connects	Properties
about.csv	3,041	Case → Topic	—
cites.csv	2,749	Case → Act	section
party_in.csv	1,309	Party → Case	role (appellant / respondent)
decided.csv	1,264	Judge → Case	—

Also included

legal-judgments.sgsnap — the same graph as a Samyama snapshot, importable in one step.

Usage

from datasets import load_dataset

cases = load_dataset("VaidhyaMegha/legal-judgments-kg", "case", revision="v1.0")
print(cases["train"].num_rows)          # 589

docker run --rm -p 8080:8080 -p 6379:6379 public.ecr.aws/f9f6l5u4/samyama-graph:1.1.0
curl -X POST localhost:8080/api/tenants -H 'Content-Type: application/json' \
     -d '{"id":"legal","name":"Legal Judgments KG"}'
curl -X POST localhost:8080/api/tenants/legal/snapshot/import -F "file=@legal-judgments.sgsnap"

// Most-cited sections — the section is on the edge, so this is one hop
MATCH (c:Case)-[r:CITES]->(a:Act)
RETURN a.name AS act, r.section AS section, count(DISTINCT c) AS cases
ORDER BY cases DESC LIMIT 5
// Indian Penal Code §302 (murder) 57 · Constitution Article 32 36 ·
// IPC §34 35 · Article 14 31 · Article 136 24

// Which judges sit together — a question about the relationship between two
// nodes of the same type, which a flat table cannot express directly
MATCH (j1:Judge)-[:DECIDED]->(c:Case)<-[:DECIDED]-(j2:Judge)
WHERE j1.name < j2.name
RETURN j1.name, j2.name, count(DISTINCT c) AS together
ORDER BY together DESC LIMIT 5
// Kurian Joseph & Rohinton F. Nariman 55

// Laws spanning the widest range of subject matter — two hops
MATCH (a:Act)<-[:CITES]-(c:Case)-[:ABOUT]->(t:Topic)
RETURN a.name, count(DISTINCT t.category) AS breadth, count(DISTINCT c) AS cases
ORDER BY breadth DESC LIMIT 5

Provenance

1. Original source — Indian Supreme Court Judgments on the AWS Open Data Registry, curated by Dattam Labs, CC-BY-4.0.

2. Intermediate — Shreyasrao/Indian-law-supreme-court-judgements-2016, revision e928c72019d6: 589 structured judgment JSONs, same licence.

3. ETL — samyama-ai/legal-judgments-graph-kg, which flattens those into node/edge tables.

4. This dataset — exported from a Samyama snapshot taken 2026-07-22 (sgsnap v2, engine v1.7.0) and flattened to CSV. Totals asserted against the snapshot header — 4,462 nodes, 8,363 edges — and re-verified after upload.

Versioning

Version	Date	Contents
v1.0	2026-08-17	Initial release — 4,462 nodes, 8,363 edges from the 2026-07-22 snapshot

Pin the tag, not main.

load_dataset("VaidhyaMegha/legal-judgments-kg", "case", revision="v1.0")

Data changes get a new tag; card corrections land on main.

Limitations

• One year only — 2016. This is not a corpus of Indian case law; it is one year of it. Citation counts are counts within these 589 judgments, not across the Court's history.
• 3 of 589 cases have no edges and an empty title — 2016-11-764-779-en.md, 2016-11-780-804-en.md, 2016-11-805-811-en.md. They appear to be source documents the extraction could not parse. Every other node in the graph is connected.
• Act names are strings, not reconciled against any statute register. "Constitution of India" and a differently-spelled variant would be distinct nodes.
• Topics are extracted, not official. The 2,291 topics and their 11 categories come from the ETL's classification of judgment text — useful for navigation, not an authoritative taxonomy.
• Party names are as written in the judgment, including titles and suffixes, so the same litigant may appear under more than one node.
• No judgment text, no outcomes, no dates beyond year and month.

Not legal advice, and not a substitute for the official record. Use the Supreme Court's own reports for anything that matters.

Verification

Totals match the snapshot header exactly, and the graph reproduces every published figure from the source repository and its engine case study:

Query	Published	This dataset
IPC §302	57 cases	57
Constitution Article 32	36	36
IPC §34	35	35
Article 14	31	31
Article 136	24	24
Dipak Misra	104 cases	104
T. S. Thakur	81	81
Kurian Joseph & R. F. Nariman together	55	55

Citation

CC-BY-4.0 requires attribution. Cite the original source and this build.

Original source — the format the AWS registry asks for:

Indian Supreme Court Judgments was accessed on DATE from
https://registry.opendata.aws/indian-supreme-court-judgments.
Curated by Dattam Labs. Licensed CC-BY-4.0.

This graph build:

Indian Supreme Court Judgments Knowledge Graph, v1.0 (4,462 nodes, 8,363 edges).
Built with Samyama Graph. https://huggingface.co/datasets/VaidhyaMegha/legal-judgments-kg
ETL: https://github.com/samyama-ai/legal-judgments-graph-kg
Source data: Indian Supreme Court Judgments, Dattam Labs, CC-BY-4.0,
             https://registry.opendata.aws/indian-supreme-court-judgments/

@misc{legal_judgments_kg_samyama,
  title        = {Indian Supreme Court Judgments Knowledge Graph},
  howpublished = {\url{https://huggingface.co/datasets/VaidhyaMegha/legal-judgments-kg}},
  version      = {v1.0},
  note         = {Derived from Indian Supreme Court Judgments, curated by Dattam Labs,
                  CC-BY-4.0,
                  \url{https://registry.opendata.aws/indian-supreme-court-judgments/}.
                  ETL: \url{https://github.com/samyama-ai/legal-judgments-graph-kg}},
  year         = {2026}
}

The three pieces

Piece	Where
Code — ETL, schema, loaders	github.com/samyama-ai/legal-judgments-graph-kg
Data — this dataset, and the raw source it derives from	here, and registry.opendata.aws
Graph — importable snapshot	legal-judgments.sgsnap in this repository