SFD: SEC Filings Dataset for Language Modeling, January 2022 to June 2025

SFD-v1 is an open, layout-faithful reconstruction of U.S. Securities and Exchange Commission (SEC) EDGAR filings into token-efficient MultiMarkdown (MMD). The dataset covers approximately 3.4 million filings from January 2022 through June 2025, produced by the SFD parser described in the associated paper. It is authored by anonymous-md and was last updated on the Hugging Face platform in May 2026.

Use Cases

• Train long-context language models based on the layout-faithful MultiMarkdown reconstruction of financial documents.
• Develop financial reasoning agents based on structured corporate disclosures from the SEC.
• Evaluate document understanding systems based on the token-efficient representation of complex filings.
• Analyze temporal trends in corporate reporting based on filings from a multi-year period.

Strengths

• Covers approximately 3.4 million SEC filings, indicating a substantial scale.
• Includes filings from a defined time range of January 2022 through June 2025.
• Data is processed into a layout-faithful, token-efficient MultiMarkdown format targeted at specific modeling tasks.
• The dataset is openly available on the Hugging Face platform.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

U.S. Securities and Exchange Commission (SEC) EDGAR system.

Collection Method

Reconstructed by the SFD parser into MultiMarkdown format.

Time Range

January 2022 through June 2025

Freshness

Last updated 2026-05-07 11:47:34; freshness should be verified.

Geography

United States (U.S. corporate filings)