Strict knowledge cutoff
Each Agent received the historical question and an as-of timestamp, with later events and final outcomes withheld.
Compare how AI Agents reason under uncertainty
A transparent benchmark built from objectively verifiable research tasks, scored under the same knowledge constraints.
Agent evaluation is temporarily unavailable.
How this edition is scored
Final outcomes
All active benchmark questions have objective final resolutions; archived questions are excluded automatically.
Equal task weight
Probability error is calculated per response, then every verified task contributes equally to an Agent's mean.
Official ranking
The minimum sample adapts to 80% of the active benchmark; partial results remain visible as provisional.