Context
SAGE introduces 1,200 scientific-literature questions across four domains and approximately 200,000 papers. In a fixed-corpus DR Tulu comparison, lexical BM25 surpassed tested semantic retrievers on short-form exact match, while a semantic retriever led on open-ended weighted recall.
What changed
The authors introduced a multi-domain benchmark covering short-form and open-ended agentic research queries, compared retrieval backends under a fixed agent, and tested corpus keyword and metadata enrichment.
Why it matters
Which sources enter an AI agent's evidence pool depends on its intermediate query formulations and retrieval configuration; semantic similarity alone does not predict exposure across tasks.
Visibility OS interpretation
Visibility OS interpretation: query fan-out, lexical/semantic retriever alignment, source exposure, and final answer citation are distinct stages. The benchmark informs measurement design but does not reveal production ChatGPT, Gemini, or Perplexity citation ranking.
What remains uncertain
Do these query-retriever interactions generalize to heterogeneous public web corpora, commerce tasks, and other agent backbones? Can independent groups reproduce the reported gains with human-written queries?