LH2-data-labs/indian-legal-records

LH2 Data — Indian Legal Records & Judgments Corpus

The most comprehensive structured Indian legal records corpus available for AI training — 267M+ case records spanning the full judicial hierarchy, paired with a pre-computed AI enrichment layer across 21M+ court orders.

---

Dataset Summary

This corpus provides structured, indexed, and partially labelled legal records from the Indian judicial system at a scale that has no public equivalent. It covers the Supreme Court of India, 25 High Courts (with all benches and jurisdictions), 600+ District and Subordinate Courts, and specialised tribunals including the National Company Law Tribunal (NCLT), Central Administrative Tribunal (CAT), and Consumer Dispute Redressal Forums. Coverage extends across every Indian state and union territory, spanning decades of judicial proceedings.

What distinguishes this corpus from raw court data scrapes or document dumps is the AI enrichment layer: every court order in the corpus is paired with pre-computed plain-language summaries, extracted key legal points, outcome classification, specific relief granted, and cited legal provisions. This is structured, production-ready training data — not raw PDFs requiring downstream extraction pipelines.

The corpus is designed for direct ingestion into AI training pipelines for foundation model pre-training, legal reasoning model fine-tuning, retrieval-augmented generation (RAG) pipeline development, and evaluation benchmark construction.

from datasets import load_dataset

# Load case records (structured metadata)
ds = load_dataset("lh2-data-labs/indian-legal-records", "case_records", split="train")

# Load AI-enriched court orders (summaries + outcomes + provisions)
ds_orders = load_dataset("lh2-data-labs/indian-legal-records", "ai_enriched", split="train")

---

Dataset at a Glance

Dimension	Scale
Total Case Records	267,000,000+
Court Orders with AI Enrichment	21,000,000+
Court Hierarchy Levels	Supreme Court · 25 High Courts · 600+ District Courts · Tribunals
States & Union Territories	All 28 states + 8 UTs
Temporal Coverage	Multi-decade (active cases refreshed daily)
Metadata Fields per Case	26+ structured fields
AI Enrichment Fields per Order	6 (summary · key points · outcome · relief · provisions · precedents)
Delivery Formats	Parquet · JSON / JSONL · PDF · Markdown · REST API
Licence	Commercial (non-exclusive and exclusive options)

---

Data Modalities

The corpus is organised into five complementary modalities, each available as a separate configuration:

case_records — Structured Case Files

Complete case metadata for every record in the corpus. Each row contains:

Field	Description
cnr	Case Number Record — the unique 16-character national identifier assigned by the Indian judiciary to every case
case_type	Standardised case type code (152+ types: Writ Petition Civil, Bail Application, Criminal Appeal, Company Petition, etc.)
case_type_desc	Human-readable case type description
registration_number	Court-assigned registration number
court_code	Unique court identifier within the national hierarchy
court_name	Full court name
court_level	Hierarchy level (Supreme Court / High Court / District Court / Tribunal)
state_code	2-letter state/UT code
district_code	District identifier within the state
bench_type	Bench composition (Single Bench / Division Bench / Full Bench / Constitution Bench)
judicial_section	Section of the court (Civil / Criminal / Writ / Appeal / Miscellaneous / PIL / Bail / Revision)
case_category	Court-assigned thematic category
case_status	Current status (71 possible statuses: Pending / Disposed / Dismissed / Admitted / Arguments / Reserved / Transferred, etc.)
filing_date	Date the case was filed
registration_date	Date the case was formally registered
decision_date	Date the case was decided (null if pending)
first_hearing_date	Date of the first hearing
last_hearing_date	Date of the most recent hearing
next_hearing_date	Scheduled date for the next hearing (null if disposed)
judges	Presiding judge(s) and bench composition
petitioners	Petitioner / appellant names
respondents	Respondent / defendant names
petitioner_advocates	Advocates representing the petitioner(s)
respondent_advocates	Advocates representing the respondent(s)
acts_and_sections	Acts of law and specific sections invoked in the case
hearing_history	Full chronological hearing history with dates and purposes
order_count	Total number of orders in the case
judgment_count	Number of final judgments
hearing_count	Total hearings conducted
ia_count	Number of interlocutory applications filed
case_duration_days	Duration from filing to decision (or to present if pending)

court_orders — Full-Text Orders & Judgments

Full-text court orders and judgments linked to their parent case record. Available as:

• Certified PDF — original court-issued order documents
• Extracted Markdown — clean, structured text extracted from PDFs for direct model ingestion
• Linked to parent case — every order joins back to the case record via cnr

ai_enriched — Pre-Computed AI Analysis Layer

The AI enrichment layer provides structured analysis for every court order in the corpus:

Field	Description
summary	Pre-computed plain-language summary of the order
key_legal_points	Extracted key legal points and reasoning chains
outcome_classification	Standardised outcome label: PETITIONER_FAVORED / RESPONDENT_FAVORED / DISMISSED / MIXED / REMANDED / PARTIAL_RELIEF
relief_granted	Specific relief granted by the court
legal_provisions_cited	Acts, sections, and connected case precedents cited in the order
order_type	Classification of order type (Interim / Final / Procedural / Directions)

litigant_records — Per-Entity Litigation History

Complete litigation history for every individual and entity that has appeared as a party in any case across the judicial hierarchy. Enables entity-level analysis, repeat-litigant identification, and cross-court litigation graph construction.

cause_lists — Daily Hearing Schedules

Real-time hearing schedules across all courts with judge, courtroom, party, and advocate information — refreshed daily. Enables temporal analysis of court workload, scheduling patterns, and judicial allocation.

---

Sample Record

{
  "cnr": "DLHC010351552024",
  "case_type": "WP_C",
  "case_type_desc": "Writ Petition (Civil)",
  "registration_number": "8371/2024",
  "court_code": "DLHC01",
  "court_name": "High Court of Delhi",
  "court_level": "High Court",
  "state_code": "DL",
  "bench_type": "Division Bench",
  "judicial_section": "APP",
  "case_status": "DISPOSED",
  "filing_date": "2024-05-29",
  "registration_date": "2024-05-30",
  "decision_date": "2026-01-30",
  "judges": ["PRATHIBA M. SINGH", "MADHU JAIN"],
  "petitioners": ["R K Tolani"],
  "respondents": ["Esic Friends Co-operative Group Housing Society Limited", "Ors."],
  "petitioner_advocates": ["AMIT SWAMI"],
  "acts_and_sections": [],
  "order_count": 2,
  "judgment_count": 2,
  "hearing_count": 1,
  "ia_count": 3,
  "case_duration_days": 611,
  "ai_enrichment": {
    "summary": "Writ petition challenging cooperative housing society decisions...",
    "outcome_classification": "PETITIONER_FAVORED",
    "key_legal_points": ["Cooperative society governance", "Member rights under state cooperative acts"],
    "relief_granted": "Directions issued to the society...",
    "legal_provisions_cited": ["Delhi Co-operative Societies Act", "Article 226 of the Constitution"]
  }
}

---

Court Hierarchy Coverage

The corpus spans the complete Indian judicial hierarchy — from the apex court down to the last subordinate court:

Supreme Court of India
├── 25 High Courts (with all benches and jurisdictions)
│   ├── Delhi High Court
│   ├── Bombay High Court (Principal Bench · Aurangabad · Nagpur · Kolhapur)
│   ├── High Court of Karnataka
│   ├── Madras High Court (Principal Bench · Madurai)
│   ├── Calcutta High Court (Principal Bench · Circuit Bench at Jalpaiguri · Port Blair)
│   ├── Allahabad High Court (Principal Bench · Lucknow Bench)
│   └── ... (20 more High Courts with all benches)
├── 600+ District & Subordinate Courts
│   ├── District Courts
│   ├── Sessions Courts
│   ├── Civil Judge Courts
│   ├── Magistrate Courts
│   └── Special Courts (CBI · NIA · POCSO · Commercial)
└── Specialised Tribunals
    ├── National Company Law Tribunal (NCLT) — 15 benches
    ├── National Company Law Appellate Tribunal (NCLAT) — 35 benches
    ├── Central Administrative Tribunal (CAT)
    ├── Consumer Dispute Redressal Forums
    ├── Income Tax Appellate Tribunal (ITAT)
    └── Customs, Excise & Service Tax Appellate Tribunal (CESTAT)

---

Pre-Training & Fine-Tuning Use Cases

• Foundation model pre-training on Indian legal text at scale — 21M+ full-text court orders provide dense legal reasoning, argumentation, and statutory interpretation signal
• Legal reasoning model fine-tuning with verified case outcomes — the AI enrichment layer provides labelled outcome data (allowed / dismissed / remanded / partial relief) for supervised training
• RAG pipeline development for legal research and case discovery — structured metadata enables precise retrieval; extracted markdown enables clean context injection
• Outcome prediction and case-law analytics — 26+ metadata fields per case enable fine-grained feature engineering for prediction models
• Evaluation benchmark construction for Indian jurisprudence — the breadth of the corpus (152 case types × 71 statuses × 11 judicial sections) enables construction of stratified evaluation sets
• Legal NER and entity extraction — parties, advocates, judges, acts, and sections are pre-structured, enabling gold-standard entity-linked training data
• Cross-jurisdictional analysis — full hierarchy coverage enables study of how similar legal questions are decided differently across courts, benches, and states
• Temporal legal analysis — multi-decade coverage with precise date fields enables study of doctrinal evolution, legislative impact, and judicial trend analysis

---

Delivery & Access

Format	Use Case
Parquet	Large-scale ML pipeline ingestion (recommended)
JSON / JSONL	Structured records with full metadata per record
PDF	Court orders as certified copies
Extracted Markdown	Clean text for direct text processing and RAG
REST API	Live querying, bulk retrieval, and real-time refresh of active cases
Custom Schema	Available on request to match buyer's training pipeline requirements

---

Dataset Statistics by Court Level

Court Level	Case Records	Orders with AI Enrichment	States/UTs Covered
Supreme Court	500K+	400K+	National
High Courts (25)	15M+	8M+	All 28 states + 8 UTs
District Courts (600+)	250M+	12M+	All 28 states + 8 UTs
Tribunals (NCLT/CAT/Consumer)	1.5M+	600K+	National

Case Type Distribution (Top 15)

Case Type	Code	Approximate Records
Writ Petition (Civil)	WP_C	18M+
Criminal Appeal	CRL_A	12M+
Bail Application	BA	15M+
Civil Suit	CS	22M+
Anticipatory Bail	ABA	8M+
Company Petition	CP	2M+
Interlocutory Application	IA	30M+
Civil Appeal	CA	5M+
Motor Accident Claims	MACT	6M+
Habeas Corpus	HCP	1.5M+
Public Interest Litigation	PIL	500K+
Special Leave Petition (Civil)	SLP_C	800K+
Execution Petition	EP	10M+
Criminal Revision	CRL_REV	4M+
Letters Patent Appeal	LPA	1M+

---

IP & Compliance

• Platform-curated legal records under exclusive commercial licensing rights — not a raw government data dump
• Full commercial and AI training rights — cleared for pre-training, fine-tuning, evaluation, and production deployment
• Indian court orders and judgments are public records of the Indian judicial system — no privacy restrictions on published court orders
• De-identification options available for sensitive case categories (POCSO, juvenile, matrimonial) where court-mandated anonymisation applies
• Compliant with India's DPDP Act 2023 for data licensing and cross-border transfer
• AI enrichment layer is proprietary — summaries, outcome classifications, and key-point extractions are original analytical work

---

Commercial Model

Flexible licensing structures designed for AI training buyers:

Licensing Option	Description
Per-court-level	Supreme Court only / High Courts only / District Courts only
Per-jurisdiction	State-level or court-specific cuts
Full corpus	All 267M+ case records across the full court hierarchy
Records only	Case metadata without AI enrichment
Records + AI enrichment	Full corpus with pre-computed summaries, outcomes, and provisions (premium)
Exclusive licence	Time-bound, per engagement

Indicative pricing available on request; negotiated per engagement based on scope, exclusivity, and intended use.

---

Key Differentiators

• Scale: 267M+ structured case records — the largest Indian legal corpus available for AI training, approximately 250× larger than the next largest public Indian legal dataset on any platform
• Full hierarchy: Supreme Court through District Courts through specialised tribunals — not just High Court judgments
• AI enrichment layer: Every order paired with pre-computed summary, key points, outcome classification, relief granted, and cited provisions — partially labelled training data, not raw text
• Rich metadata: 26+ structured fields per case enable fine-grained filtering by court, case type, jurisdiction, judge, bench composition, acts invoked, or temporal range
• Real-time refresh: Active cases are refreshed from source, ensuring the corpus stays current with ongoing proceedings
• Production-ready formats: Parquet-first delivery with JSON/JSONL and REST API access — designed for ML pipeline ingestion, not manual research

---

Comparison with Public Alternatives

Dataset	Records	Courts	AI Layer	Orders	Access
This corpus	267M+	SC + 25 HC + 600 DC + Tribunals	21M+ enriched	Full text + PDF	Commercial
opennyaiorg/InJudgements	~12K	SC + 16 HC + 2 DC	None	Text only	Apache-2.0
debkanchan/SC-judgements	~37K	SC only	None	Text only	Public
Exploration-Lab/IL-TUR	~35K tasks	SC + HC + DC	Partial (task labels)	Excerpts	CC-BY-NC-SA
vihaannnn/SC-Chunked	~21K chunks	SC only	None	Chunked text	MIT

---

Limitations and Considerations

• Jurisdictional scope: This corpus covers the Indian judicial system exclusively. It does not include foreign court decisions, international arbitration awards, or quasi-judicial proceedings outside the formal court hierarchy.
• Language: The majority of orders are in English (the official language of the higher judiciary). District Court orders in vernacular languages (Hindi, Marathi, Tamil, Bengali, etc.) are available but AI enrichment quality may vary for non-English orders.
• Temporal completeness: While the corpus spans decades, the density of digitised records increases significantly from 2010 onward, reflecting the phased digitisation of the Indian court system.
• AI enrichment accuracy: The AI enrichment layer is generated through automated pipelines and has not been individually verified by legal professionals for every order. Users should treat enrichment fields as high-quality labels suitable for training, not as authoritative legal analysis.

---

Citation

@dataset{lh2_indian_legal_2025,
  title={LH2 Data — Indian Legal Records & Judgments Corpus},
  author={LH2 Data Labs},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/lh2-data-labs/indian-legal-records},
  note={267M+ structured case records across the full Indian judicial hierarchy with AI-enriched court orders}
}