Real-world scientific literature discovery is far from shallow keyword retrieval; it demands rigorous logical reasoning across domain-specific constraints.
SciResearchBench formalizes scientific search into two complementary task paradigms:
- Deep Research: Pinpointing a unique target paper via progressive multi-step probing and full-text verification, or determining that no qualified paper exists.
- Wide Research: Exhaustively compiling the complete set of papers satisfying strict scientific conditions (averaging 9.23 valid papers per query).
