rtarun789/indian-court-decisions

Indian Court Decisions

A large-scale dataset of Indian court decisions with full text, metadata, and outcome labels covering the Supreme Court of India and 25 High Courts (1950–2026).

Dataset Summary

Config	Train	Validation	Test	Total
high_courts	11,682,776	1,459,319	1,457,934	14,600,029
supreme_court	40,044	4,990	5,019	50,053
Total				14,650,082

This is one of the largest publicly available legal NLP datasets, containing over 14.6 million court decisions with extracted full text. The dataset covers 75 years of Indian jurisprudence across all major courts.

High Courts Coverage

Court	Records
Madras High Court	1,547,108
Patna High Court	1,537,587
High Court of Punjab and Haryana	1,295,096
Allahabad High Court	1,279,159
Bombay High Court	1,263,189
High Court of Kerala	1,031,406
High Court of Rajasthan	921,771
High Court of Karnataka	896,690
High Court for State of Telangana	885,224
High Court of Madhya Pradesh	528,650
Orissa High Court	525,161
High Court of Chhattisgarh	520,620
High Court of Jharkhand	392,123
High Court of Gujarat	375,836
High Court of Delhi	311,666
+ 10 more High Courts	...

Schema

Field	Type	Description
cnr	string	Case Number Record (unique identifier)
source	string	hc (High Court) or sc (Supreme Court)
court_code	string	Numeric court code
court_name	string	Full court name
bench	string	Bench composition
year	int16	Year of the decision
full_text	string	Full text of the judgment (min 1,000 chars)
text_length	int32	Character count of full text
title	string	Case title
judge	string	Presiding judge(s)
petitioner	string	Petitioner name (SC only)
respondent	string	Respondent name (SC only)
decision_date	string	Date of the decision
disposal_nature	string	Raw disposal outcome
disposal_nature_normalized	string	Normalized outcome label
case_type	string	Extracted case type prefix (WP, CRA, BA, FA, etc.)

Outcome Labels (disposal_nature_normalized)

Label	Description
dismissed	Case dismissed (on merits, as infructuous, not pressed)
disposed	Disposed of (general resolution)
allowed	Appeal/petition allowed
partly_allowed	Partly allowed
withdrawn	Withdrawn by petitioner
closed	Case closed
bail_granted	Bail granted
bail_rejected	Bail rejected
quashed	Order/FIR quashed
transferred	Case transferred
settled	Settled between parties
ordered	Court order issued
other	Other outcomes

Splits

Splits are deterministic based on MD5 hash of the CNR identifier (80/10/10 train/validation/test). This ensures reproducibility without requiring in-memory shuffling of 14M+ records.

Data Source

Source: AWS Open Data Registry -- Indian Court Decisions dataset by Dattam Labs. Original data collected from eCourts (ecourts.gov.in) and the Supreme Court of India.

Languages

Judgments are primarily in English, with some decisions in Hindi, Tamil, Telugu, Kannada, Marathi, Bengali, Malayalam, Gujarati, Odia, and Punjabi.

Use Cases

• Judgment prediction: predict case outcome from facts and arguments
• Legal text classification: case type, court, jurisdiction
• Temporal analysis: how judicial reasoning evolves over 75 years
• Cross-jurisdictional comparison: patterns across 25 High Courts
• Legal language modeling: pretraining or fine-tuning legal LLMs
• Information retrieval: legal search and citation prediction

Paper

This dataset is introduced in:

Multi-Legal-Bench: Evaluating LLMs on Legal Reasoning Across Jurisdictions, Languages, and Legal Traditions Volodymyr Ovcharov (2026). arXiv:2605.29738 [cs.CL]

Citation

@article{ovcharov2026multilegalbench,
  title={Multi-Legal-Bench: Evaluating LLMs on Legal Reasoning Across Jurisdictions, Languages, and Legal Traditions},
  author={Ovcharov, Volodymyr},
  journal={arXiv preprint arXiv:2605.29738},
  year={2026},
  url={https://arxiv.org/abs/2605.29738}
}

License

CC-BY-4.0