Evals Router

Route LLM/RAG eval work to the smallest proof-producing path.

When to use

Use for eval coverage, failed traces, judge prompts, labels, RAG evidence, synthetic cases, or review tooling. Ask one question only when target, traces, labels, scorecard, or desired artifact is missing.

Required inputs

Need the user goal plus any traces, labels, scorecard, prompt, retrieval evidence, or target artifact.

Routes

Use eval-audit, error-analysis, write-judge-prompt, validate-evaluator, generate-synthetic-data, evaluate-rag, or build-review-interface.

Faithfulness guardrails require owned sources, sentence-level verdicts, structured judge outcomes, labeled examples, exact pass references, and calibration.

Deliverables