Experimental utilities for evaluating and guarding retrieval behavior in RAG systems.
The active Python package and full project README live in retrieval-guard/.
RAG systems can fail quietly when retrieved context looks plausible but does not actually support the user's question. retrieval-guard explores lightweight benchmarking and verification ideas for catching retrieval regressions and near-miss failures before they flow into downstream generation or agentic workflows.
Experimental prototype. Not a production library.