NormasTCU

Overview

NormasTCU iis a dataset for Legal Information Retrieval (LIR) in Brazilian Portuguese composed of normative documents from the Brazilian Federal Court of Accounts (Tribunal de Contas da União - TCU), along with queries and human-annotated relevance judgments.

The dataset includes:

• 14,469 legal documents (normative acts);
• 46 queries;
• 812 judge query-document pairs derived from 3,048 human annotations with 3-level graded relevance.

Usage

To use the dataset, load que files docs.csv, query.csv, and qrel.csv:

import pandas as pd

docs = pd.read_csv("docs.csv")
queries = pd.read_csv("query.csv")
qrels = pd.read_csv("qrel.csv")

Dataset Structure

The NormasTCU dataset consists of three main files:

NormasTCU/
│── doc.csv               # Corpus
│── query.csv             # Queries
│── qrel.csv              # Relevance judgments for each query
│── raw_human_eval.csv    # Anonymized relevance judgments from each human annotator (before aggregation)

Documents (doc.csv)

The doc.csv file contains 16,045 legal documents from the TCU's selected jurisprudence. Each row represents a legal document and includes the following fields:

Fields of a Document in the Corpus

Field	Description	Type	Missing values
KEY	ID	Text	--
UNIDADEBASICAAUTORA	Superior authoring organizational unit	Text	--
ORIGEM	Source of the normative act	Text	--
NUMNORMA	Document number	Text	--
ANONORMA	Document year	Text	--
TIPONORMA	Document type	Text	--
TITULO	Document title	Text	--
ASSUNTO	Subject / summary	Text	7,191 (49.70%)
TEXTONORMA	Full text	HTML Text	5 (0.03%)
DATAINICIOVIGENCIA	Effective start date	DD/MM/YYYY	--
DATAFIMVIGENCIA	Effective end date	DD/MM/YYYY	13,629 (94.19%)
SITUACAO	Expressly revoked or in force	Text	--
TEXTOANEXO	Annex of the document	HTML Text	13,237 (91.49%)
TEMA	Whether the document is about external control or management	Text	3,686 (25.48%)
TAGSVCE	Tags	Text	3,849 (26.60%)
NORMARELACIONADA	Related documents of this dataset	Text	9,918 (68.55%)

Queries (query.csv)

The query.csv file contains 46 standardized queries. Each row in the file contains the following fields:

Field	Description
KEY	Unique query identifier
TEXT	Query text

Relevance Judgments (qrel.csv)

The qrel.csv file contains relevance assessments for query-document pairs.

Each row in the file contains the following fields:

Field	Description
QUERY_ID	Query identifier
DOC_ID	Document identifier
SCORE	Relevance score (0 = irrelevant, 1 = partially relevant, 2 = relevant)

Citation

If you use the NormasTCU dataset, please cite:

@misc{normastcu2026,
  author    = {Fernandes, Leandro Carísio and 
               de Castro, Marcos Vinícius Borela and 
               Ribeiro, Leandro dos Santos and 
               da Silva Pacheco, Leonardo Augusto and 
               de Oliveira Sandes, Edans Flávius},
  title     = {{NormasTCU}}
}

Contact

• Leandro Carísio Fernandes
  Câmara dos Deputados, Brasília, Brazil
  Email: carisio@gmail.com

• Marcos Vinícius Borela de Castro
  Tribunal de Contas da União (TCU), Brasília, Brazil
  Email: borela@tcu.gov.br

• Leandro dos Santos Ribeiro
  Tribunal de Contas da União (TCU), Brasília, Brazil
  Email: leandro.santos.r@gmail.com

• Leonardo Augusto da Silva Pacheco
  Tribunal de Contas da União (TCU), Brasília, Brazil
  Email: leonardo3108@gmail.com

• Edans Flávius de Oliveira Sandes
  Tribunal de Contas da União (TCU), Brasília, Brazil
  Email: edansfs@tcu.gov.br

Acknowledgments

We would like to thank the Brazilian Federal Court of Accounts (TCU) for providing the documents and supporting this research.