FiledFact-100K: 100,000 SEC Financial Facts Grounded in Filing Text

FiledFact-100K provides 100,000 pairs linking XBRL-tagged financial facts to the exact character spans where they appear in SEC filing text. Each record includes a passage of clean filing text and a fact with its XBRL concept, period, value, unit, and decoded segment dimensions. The dataset was created by StockAlloy and was last updated on July 7, 2026.

Use Cases

• Train models for financial named entity recognition based on XBRL concepts and values.
• Develop information extraction systems for SEC filings based on the text-to-fact alignments.
• Build fact-checking or verification tools for financial disclosures based on the value-verified fact pairs.
• Fine-tune language models on domain-specific financial text from statements, notes, and MD&A sections.

Strengths

• Contains 100,000 precisely aligned text-fact pairs.
• Every record includes a verified character-level span linking the XBRL fact to its displayed text in the filing.
• Each fact is annotated with XBRL metadata including concept, period, value, unit, and segment dimensions.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known (100,000), but the total size and file formats are unspecified.
• The license is unknown, which may restrict usage.

Provenance

Source

SEC filings

Collection Method

Likely involves parsing and aligning XBRL data with filing text.

Freshness

Last updated 2026-07-07 23:12:55; freshness should be verified.