Evaluate human-guided focal selection

Evaluate the human-guided focal-selection configuration of JuryFlow through a user study comparing human selection with entropy-based selection in terms of accuracy and effort.

Background

JuryFlow identifies high-disagreement claims using verdict entropy and can either select a focal claim automatically or present ranked candidates to a human. All experiments in the paper use the automatic entropy-ranking configuration, so the empirical contribution of a human structural guide remains unmeasured. The unresolved evaluation concerns whether human selection improves accuracy or changes the effort required relative to the entropy proxy.