Delegation Intelligence: A Framework for Disentangled Deep Search Diagnosis

A recent study presents Delegation Intelligence, a framework designed to assess the deep search capabilities of AI agents. The researchers contend that existing end-to-end accuracy metrics mix together retrieval, comprehension, verification, and decision-making regarding tool use. They break down Delegation Intelligence into two components: Search Decision-Making (identifying when information is lacking and determining the timing and method of searching) and Information Synthesis and Verification (collecting evidence, evaluating source credibility, and synthesizing information in challenging environments). To facilitate clear and reproducible measurement, a controllable synthesis pipeline has been created. This paper can be found on arXiv with the ID 2607.23524.