BRIGHT-Pro
Reasoning-intensive retrieval quality on real-world StackExchange questions, scored against multi-aspect gold evidence. Each query is paired with a long-form reference answer whose cited passages collectively cover several reasoning aspects, so retrievers are measured on surfacing an aspect-diverse evidence set rather than a single relevant passage. Extends BRIGHT with these aspect annotations for agentic search settings.
Reference paper →Languages 1
Tasks 7
Task Types 1
Models 0