Summary

• add general RAG answer eval primitives for context quality, answer support, citations, abstention, and failure diagnosis
• add external score normalization and row exporters for Ragas, DeepEval, TruLens, RAGChecker, and custom evaluators
• add generic KB quality scoring for citation coverage, source freshness, source-backed pages, duplicates, and lint/validation failures

Verification

• pnpm exec biome check src tests
• pnpm exec tsc --noEmit
• pnpm exec vitest run tests/rag-eval.test.ts tests/rag-improvement-loop.test.ts tests/retrieval-eval.test.ts
• pnpm exec tsup
• pnpm exec vitest run