CoverSynth: Coverage-Guided Generation of Retrieval Evaluation Sets
CoverSynth makes coverage and evidence quality explicit construction objectives, producing corpus-grounded evaluations that reveal retrieval headroom and differences between agent configurations that reveal unseen failure modes such as an inability of agents to handle underspecified or ambiguous requests.
Anahita Sharma, Apaar Shanker, Rishav Chakravarti et al.
· 0 citations