Context
An original work-in-progress preprint introduces IndexAct, an index-native interface allowing search agents to manipulate persistent candidate sets using lexical conditions and set operations while retrieving text passages separately. The authors report improvements over evaluated baselines on five benchmarks.
What changed
IndexAct exposes candidate-set state and statistics before passage delivery, rather than returning matching text for every refinement step. The authors report higher evidence coverage with smaller average live context on BrowseComp-Plus than terminal-based corpus interfaces.
Why it matters
Agentic discovery can depend on how an agent narrows candidate sources and when it chooses to inspect evidence, not only on which documents a search index initially matches.
Visibility OS interpretation
Visibility OS interpretation: Candidate refinement, evidence exposure and answer citation are distinct operations. Index-native state may improve agent search efficiency in tested settings; no production ChatGPT, Google, Claude or Perplexity behavior is established.
What remains uncertain
Does the approach generalize to open-web indexing and changing corpora? What are the computational trade-offs and false-negative risks from lexical candidate refinement? Does greater evidence coverage improve citations and answer faithfulness?