Benchmark RadarTCR-Bench
TCR-Bench is a diagnostic benchmark for table content-level answerability in RAG, focusing on sibling tables with similar schemas but content differences. It evaluates dense retrievers' ability to identify answerable tables.
- Released
- 2026-07-20
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
This benchmark probes the semantic-answerability gap in retrieval for table RAG, showing that semantic relevance does not guarantee answerability. It provides diagnostic insights into retriever limitations and the potential for answerability-aware reranking, but is primarily for research findings.
Motivation
Tables are a critical knowledge source in retrieval-augmented generation (RAG), but a retrieved table may lack sufficient evidence to answer a query, a property we call answerability.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.