Skip to content

DecisionBench

DecisionBench is the evaluation ecosystem for decision models: systems that turn documents or structured state into calibrated decisions over runtime-defined candidates.

The frozen 1.0 release contains 23,900 examples, nine use-case families, three output primitives, and candidate sets from 2 to 255. The runtime provides common adapters, scoring, raw artifact contracts, reviewed result records, and an interactive leaderboard.

Start here

  1. Install DecisionBench.
  2. Select tasks or views.
  3. Run an evaluation.
  4. Load or submit results.
  5. Explore the leaderboard.

The canonical dataset is published at Hanno-Labs/decision-bench.