Observed arrival · 2026-10-20
Tickmark Bench turns SEC filings into agent tests
A benchmark pack asks AI agents questions grounded in SEC EDGAR XBRL filings, with answers checked against filing-derived references.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- AI labs and RL environment vendors
- Worth noticing
- Core's official score is passed/149; the site says solve rates are not yet published.
Field notes
The current Core set is described as 149 verified tasks in the real_v1 batch; its official score is passed/149, produced by grade.py. Numeric tasks are checked against filing-derived values within stated tolerances, while claim checks score only the yes-or-no verdict—the explanation is not scored. The downloadable 20-task sample uses an earlier format and is explicitly excluded from Core; the page does not publish task-type counts or solve rates.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
◎NicheUnusually specific use
One card from the complete issue