Benchmark RadarMemoryDocDataSet
MemoryDocDataSet evaluates joint conversational memory and long-document reasoning through 50 synthetic micro-worlds, each with personas, temporal event graphs, real legal documents, multi-session conversations, and QA pairs. Questions are categorized by reasoning type, with Hybrid questions requiring navigation of conversation history to locate the relevant document and extract answers.
- Released
- 2026-06-03
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Existing benchmarks evaluate conversational memory or document reasoning separately, leaving a gap in measuring integrated performance. MemoryDocDataSet provides a controlled, synthetic environment to assess systems on tasks that require both capabilities, useful for developing and comparing architectures that unify conversation memory with long-document retrieval.
Motivation
AI systems increasingly need to combine two demanding capabilities: navigating multi-session conversation history and performing deep reading comprehension within long documents.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.