5,698 passages from SEC filings contain 101,899 financial facts, each grounded to its exact character span in the original text. Every numeric span is labeled as either a target fact or a non-target, and facts are linked to their original numbers. The dataset, created by StockAlloy, was last updated on July 18, 2026.
Use Cases
- Train named entity recognition models to extract financial concepts based on the grounded fact annotations.
- Develop relation extraction systems linking financial values to their concepts and periods based on the structured fact tags.
- Benchmark models for span detection and numeric grounding in financial text based on the character-level annotations.
- Analyze the distribution and presentation of financial facts across different sections of SEC filings like MD&A and financial statement notes.
Strengths
- Contains 101,899 precisely annotated financial facts across 5,698 passages.
- Provides character-level grounding for every fact and labels all other numeric spans as non-targets.
- Links extracted facts to the exact numbers in the original SEC filings for verification.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Last updated 2026-07-18 21:56:15; freshness should be verified.
Provenance
- Source
- SEC filings.
- Collection Method
- Passage extraction and fact annotation from financial filings, likely involving automated and manual processes.
- Freshness
- Last updated 2026-07-18 21:56:15.