Back to feed
arXiv cs.CL
arXiv cs.CL
7/31/2026
LayerRAG-Bench: A Cross-Layer Reliability Benchmark for Agentic Retrieval-Augmented Generation

LayerRAG-Bench: A Cross-Layer Reliability Benchmark for Agentic Retrieval-Augmented Generation

Short summary

LayerRAG-Bench introduces a cross-layer reliability benchmark for agentic RAG systems, testing 240 tasks across 8 enterprise domains with 9 fault scenarios and nearly 39K records from OpenAI, Anthropic, and Gemini models. Schema normalization fixes schema-drift but fails to recover stale evidence, missing tool output, denied permissions, or wrong-session context. The key finding: reliability interventions should be evaluated layer-specifically, not treated as universal fixes.

  • •Benchmark covers 8 enterprise domains, 240 tasks, 38,880 records across 9 models
  • •Schema normalization raises schema-drift success from 0 to 0.913 but doesn't fix other fault layers
  • •Layer-specific evaluation principle: credit interventions for their target layer only

Generated with AI, which can make mistakes.

#research-breakthrough#ai-agents#ai-tools
Read full article at arXiv cs.CL

Is this a good recommendation for you?

Comments

Failed to load comments. Please try again.

Explore more

Relay-Bench: Evaluating LLMs on Multi-Domain Reasoning Chains
arXiv cs.CL
RAG Explained: Retrieval-Augmented Generation pipeline for grounded LLM responses
Dev.to
The original title is about RAG evaluation with RAGAs metrics. Let me rewrite it to be punchy and specific while preserving key facts.
Dev.to
RAG in production: the failure modes nobody warns you about
Dev.to