The query that was posed to the RAG pipeline under evaluation.
The answer generated by the RAG pipeline under evaluation.
The document chunks retrieved by the RAG pipeline and injected into the LLM's prompt. Presented to the judge in numbered order ([Context 1], [Context 2], ...). Ordering matters for contextPrecision scoring.
OptionalgroundThe expected / reference answer for this question.
Only required by contextRecall -- other metrics ignore this field.
When absent, contextRecall returns skipped: true for this sample.
Input passed to each metric's
score()function.Mirrors one row of the evaluation dataset — all fields come directly from the RagSample passed to
evaluate(). The metric implementations receive this object and use whichever fields are relevant to their scoring logic.