Learning to Evaluate Before Improving: Automatic Rubric Induction for Automatic Research Agents
AutoSciRub has research agents write an executable task rubric before they start the work.
The framework turns underspecified research instructions into atomic goals, grounds them in literature and visible data, and converts them into verifiable criteria. It then uses that rubric to steer execution, check unmet criteria, and revise reports and supporting artifacts. The paper reports gains across ResearchClawBench and a 20-task subset of AstaBench E2E Discovery, including a 16.8-point average improvement on the latter. The authors mark the paper as work in progress. HF Daily Papers' note
The framework turns underspecified research instructions into atomic goals, grounds them in literature and visible data, and converts them into verifiable criteria. It then uses that rubric to steer execution, check unmet criteria, and revise reports and supporting artifacts. The paper reports gains across ResearchClawBench and a 20-task subset of AstaBench E2E Discovery, including a 16.8-point average improvement on the latter. The authors mark the paper as work in progress. HF Daily Papers' note
score 5