StockAlloy — grounded financial data from SEC filings

We turn SEC XBRL filings into span-grounded, value-verified data: every fact is linked to the exact characters that display it in the filing, with a deep link to the source document.

Datasets

• FiledFact-100K — 100K SEC financial facts, each grounded to its exact filing text and deep-linked to the source.
• FiledFact-Passages — the passage-complete companion: every filer-tagged fact in a filing passage (5,698 passages · 101,899 facts · 4,878 companies), with non-target numbers labeled so a model learns what not to extract.

Models

• FiledFact-Qwen3-8B — Qwen3-8B fine-tuned for SEC fact extraction + grounding. Reads a filing passage, returns the XBRL-aligned facts as structured JSON.

Roadmap

• FiledFact-Bench — a static benchmark for financial fact extraction

Contact

• data@stockalloy.com · stockalloy.com/datasets