Statskontoretdatalabb/Regeringsuppdrag

Swedish government assignments

This release contains a sample of 100 publications and 101 PDF files, with extracted text, Markdown and Docling JSON. It is not the complete collection of Swedish government assignments.

Swedish government assignments (regeringsuppdrag) with publication details, receiving organisations, extracted text and original PDF files. The source is Regeringen.se. Text and structured content are extracted using Docling.

Structure

Each row represents one file. Files belonging to the same publication share a publication_id. Original PDFs are stored separately under pdfs/, so you can read the extracted content from regeringsuppdrag.parquet without downloading PDFs.

The dataset includes decisions, appendices and other associated files without classifying their roles. Use file titles and the original PDFs to assess their content; sequence indicates source order, not importance.

Fields

Field	Description
id	Unique file ID.
publication_id	ID shared by files from the same publication.
title	Publication title.
author	Publishing organisation: Regeringskansliet.
published	Publication date.
registry_numbers	List of registry numbers (diarienummer).
recipients	List of receiving organisations.
context_url	Link to the publication page.
content_title	Title of the individual file.
sequence	File order on the source page.
file_name	Original filename.
url	Link to the original PDF.
pdf_path	Path to the PDF within this dataset.
pdf_md5	MD5 checksum for comparing PDF file contents.
text	Extracted plain text.
markdown	Extracted content in Markdown format.
json	Structured Docling content, stored as a JSON string.
docling_version	Docling version used for extraction, when available.

Missing values are stored as null. registry_numbers and recipients are lists; use json.loads() to read the structure inside json.

Read the data

After downloading regeringsuppdrag.parquet, you can read selected columns with Python and PyArrow:

import pyarrow.parquet as pq

table = pq.read_table(
    "regeringsuppdrag.parquet",
    columns=["title", "recipients", "markdown"],
)
print(table.slice(0, 1).to_pylist())

To check the original document, follow url or download the file at pdf_path. The Hugging Face split is named train for loading purposes; it does not imply that the data is a curated training set.

Limitations

Text extraction can contain errors, especially in tables and scanned documents. Recipient matching may also be imperfect. Registry numbers omit the source's “delvis” qualifier. Check the original PDF when these details matter. Coverage depends on the available source material; the dataset does not guarantee that every Swedish government assignment is included.