CoolFace
Datasetpublic

SAFE-Action/us-science-policy-bills

US Science Policy Bills A stance-annotated corpus of 8,191 public-health-related bills from the United States federal Congress and 41 state legislatures, collected and classified by SAFE Action (Science and Freedom for Everyone Action Fund), a 501(c)(4) social welfare organization building open-source democratic tools for science-based policy. To our knowledge this is the first openly published, stance-annotated dataset of state-level public health legislation. Raw legislative… See the full description on the dataset page: https://huggingface.co/datasets/SAFE-Action/us-science-policy-bills.

sourceHugging Facecc-by-4.0updated 2mo agoView on Hugging Face
0likes9downloads
Dataset Card

US Science Policy Bills

A stance-annotated corpus of 8,191 public-health-related bills from the United States federal Congress and 41 state legislatures, collected and classified by SAFE Action (Science and Freedom for Everyone Action Fund), a 501(c)(4) social welfare organization building open-source democratic tools for science-based policy.

To our knowledge this is the first openly published, stance-annotated dataset of state-level public health legislation. Raw legislative data is available from sources like LegiScan and Open States; the annotation layer here (stance, impact, category, plain-language summaries) is what SAFE Action adds.

What's in it

One JSON record per bill (bills.jsonl):

fielddescription
billId, billNumberidentifiers
title, summaryofficial title and plain-language summary
state, levelstate code; federal or state
billType, status, isActive, lastAction, lastActionDateprocedural state
category, subjectstopic classification
stanceSAFE Action's annotation: support, oppose, or monitor
impactestimated impact: high, medium, low
sponsorssponsoring legislators (public officials, public record)
committee, urlcommittee and link to the official bill page

Label distribution (2026-07-24 snapshot): monitor 8,036 / oppose 136 / support 19. The corpus is a live monitoring feed, so labels are heavily imbalanced toward monitor; treat support/oppose as the high-signal minority classes.

Methodology

Bills are ingested daily by an automated crawler covering all 50 states and Congress, filtered to public-health relevance, then classified through a three-stage pipeline: keyword scanning, LLM-based analysis, and an adversarial review pass in which a second model challenges the first model's classification before a label is accepted. Full pipeline source: SAFE-Action/SAFE-Action-Website (GPLv3), crawler/.

Licensing

  • —Bill text, titles, and legislative metadata are US government works (public domain).
  • —SAFE Action's annotations (stance, impact, category, summaries) are released under CC-BY-4.0. Please attribute "SAFE Action (scienceandfreedom.com)".

Intended uses

Training and evaluating civic/legislative text classifiers, policy-tracking research, journalism, and advocacy tooling. This dataset reflects the annotating organization's science-based-policy perspective; the stance labels encode that editorial standard and should be treated as such in research use.

Updates

Snapshot dated 2026-07-24. The underlying tracker updates daily; periodic snapshot refreshes are planned.

Citation

@dataset{safeaction2026bills,
  title  = {US Science Policy Bills},
  author = {{SAFE Action (Science and Freedom for Everyone Action Fund)}},
  year   = {2026},
  url    = {https://huggingface.co/datasets/SAFE-Action/us-science-policy-bills}
}