Finance reasoning
Financial QA 10K
Locate and interpret disclosures in 10-K narrative sections (MD&A, risk factors, business).
As of 13 August 2026. Conditional pack. Hosted 10-K QA mirrors. License not verified as commercial-clear. No cited public figure is installed.
Financial QA 10K asks whether a model can locate and interpret 10-K narrative, not compute a ratio from a handed table. Answer accuracy, evidence-span support, and numeric match where the gold is numeric stay separate. They are not averaged into one ranking.
The hosted Hugging Face card is not an approved commercial grant. Recheck terms before a run. No official TAT-QA-style table for GPT-5.6 or Fable 5 is cited here.
- Scale
- About 7,000 QA pairs. Pin the hosted revision at acquisition.
- Provenance
- Commonly mirrored as virattt/financial-qa-10K. No explicit commercial data license verified here.
- Contamination
- High. Public Hugging Face mirrors. Recheck license before a commercial run.
- Cannot establish
- A 10-K research-product score, AlphaSense performance, or economic value. Hosted license is not verified as commercial-clear.
Metrics, kept separate
- Answer accuracy
- Evidence-span support
- Numeric match where the gold is numeric
No cited public figure is installed for this pack. Method, scale, and gates still apply. A missing chart is not a pass rate.
What the figures bound
No cited public figure is installed. Use the pack, after rights clearance, as a 10-K narrative diagnostic. Do not treat a hosted mirror as AlphaSense or as official TAT-QA exact match.
Absence of a chart is the honest state until a frozen, rights-cleared replay exists.
What this page does not conclude
- A commercial run before the hosted license is approved
- AlphaSense or filing-research product performance
- GPT-5.6 or Fable 5 on this pack
- An unweighted composite ranking