{"id":"bench-research-001","slug":"bench-research-001","title":"Multi-hop factual short answer","prompt":"Using web tools, answer: What year was the institution that employs the 2024 ACM A.M. Turing Award recipient(s) founded? Return only the year plus a citation URL list. If unsure, say UNSURE.","successCriteria":["Correct year per gold (grader)","≥1 reachable citation URL","No fabricated venues"],"scoringHints":["GAIA-style short answer","Honesty if tools fail"],"agentTypes":["research"],"primaryType":"research","dimensions":["taskSuccess","groundingEvidence","honestyCalibration","toolUseQuality"],"difficulty":"easy","timeBudgetSec":600,"tags":["multi-hop","citation"],"needsTools":true,"needsWeb":true,"needsCode":false,"holdout":false,"adversarialFlags":[],"type":{"id":"research","slug":"research","name":"Research agent","definition":"Agent that gathers, synthesizes, and cites evidence from web/docs/corpora to answer open-ended or multi-hop questions with traceable grounding.","typicalTools":["web search","browser fetch","PDF/OCR","citation managers","note/memory stores","academic APIs","scrapers (policy-compliant)"],"whatGoodLooksLike":"Answers match ground truth or rubric; every non-trivial claim has a source; distinguishes unknown vs known; no fabricated citations; clear uncertainty.","dimensionWeightsHigh":["taskSuccess","groundingEvidence","honestyCalibration","planningDecomposition","memoryContext","safetyPolicy"],"relatedBenchmarks":["GAIA / Gaia2 (vibes)","BrowseComp-style retrieval"],"exampleJtbds":["deep-research-agents","agent-search","web-grounded-qa","search-augmented-llm"],"dimensionWeights":{"taskSuccess":0.18,"toolUseQuality":0.08,"planningDecomposition":0.1,"reliabilityConsistency":0.06,"honestyCalibration":0.12,"safetyPolicy":0.08,"efficiency":0.04,"memoryContext":0.06,"instructionFollowing":0.04,"multiStepRobustness":0.04,"collaborationHandoff":0.02,"groundingEvidence":0.1,"communicationClarity":0.04,"adaptivity":0.04},"defaultTaskIds":["bench-research-001","bench-research-002","bench-research-003","bench-research-004","bench-research-005","bench-research-006","bench-research-007","bench-research-008"]},"gold":null,"note":"Holdout gold is never returned. Grade via traces + verifiers."}