Human-in-the-Loop Nugget Annotation for Accountable LLM-as-a-Judge Evaluations
For researchers and practitioners evaluating AI systems, this work addresses the challenge of reliable human oversight in LLM-based evaluation, though it is an incremental design contribution.
The paper presents a human-in-the-loop annotation tool where humans identify key information (nuggets) and LLMs match them to outputs, improving accountability in LLM-as-a-judge evaluations. The approach aims to reduce anchoring bias and support high-variance labeling.
Evaluating AI/Agentic system outputs reliably requires human judgment, but how one incorporates the human determines whether one gets a real quality signal or expensive theater. The common approaches either accidentally anchor human experts (leading to rubber-stamping) or leave them unsupported in high-variance labeling tasks. We present a prototype annotation tool that implements a different division of labor: humans identify what information matters (nuggets), while LLMs handle high-volume matching of nuggets to system outputs. This plays to each party's strengths while maintaining genuine human oversight. We describe the three-phase workflow, key design decisions, and how exported nugget banks integrate with automated judges.