Benchmark¶

DecisionBench 1.0 is a named collection of 43 registered tasks containing 23,900 English examples over nine families and three primitives. Candidate sets range from 2 to 255. The release combines applied decision tasks with a reasoning track while keeping task identity visible for filtering and separate aggregates.

Those first-release tasks share the consolidated dataset Hanno-Labs/decision-bench at revision b7c8107e01ecb1aee7c7eaf5caee4a3ba9f59443. New tasks may own independent pinned datasets; a benchmark composes task IDs rather than requiring one canonical data repository.