Visual agents evaluated for accuracy using budget-paired counterfactual audits

Choosing What to Observe: Budget-Paired Counterfactual Audits for Credible Evaluation of Visual Agents

Artificial Intelligence

Summary

Visual agents that read and answer questions about documents have to pick which parts to look at because they have limited time or attention. The authors propose a new way to test these agents fairly by keeping some information fixed and changing only the budget or task rules, allowing clear measurement of how much is lost due to decision steps versus final answers. This approach separates the effects of planning where to look from how the agent answers once it stops looking. The method was tested on multiple cases to better understand the agent's behavior and accuracy.

What this means in practice

Authors

Ruiqi Liu, Jinming Li, Mo Wang, Yang Lu, Yanlan Kang

Abstract

Visual agents that read dense documents must ration a finite observation budget across candidate regions, yet answer accuracy conflates where an agent looks, how it adapts to the reads that remain, and how it an?swers once observation stops. We introduce budget-paired counterfactual audits, an evaluation protocol that holds public evidence and region bindings fixed while perturbing only the remaining budget or de?clared task parameters, and credits every acquired history with exact continuation values. The resulting accounting decomposes lost expected accuracy into stepwise acquisition loss and terminal decision loss, separating planning from fixed-role heuristics. Across forty paired routing anchors, five open-weight