Agent Evaluation

FrontierChallenge: Evaluating Scientific Workflow Completion featured image

FrontierChallenge: Evaluating Scientific Workflow Completion

97 scientific workflows across six domains, evaluated on complete artifact delivery. The best reported system finished 20.6% of tasks despite strong partial scores.

Liangcai Su
Apodex Discovery: Reality Benchmarks and Environments for Evaluating and Building Discoverative Artificial Intelligence featured image

Apodex Discovery: Reality Benchmarks and Environments for Evaluating and Building Discoverative Artificial Intelligence

Executable research environments and HDS6 process verification, with concrete cases showing how evidence-based feedback can repair an investigation.

Brian Wang