3.4Standard
Retrieval looks good but final answers are wrong.
Separate failure layers.
Choose two evaluations.
× Incorrect
Measure retrieval recall, ranking, authorization, evidence support, omissions, safety, latency, and cost on the same question set.
Detailed explanation
CorrectMeasure retrieval recall, ranking, authorization, and generation support separately
Failure layers are separated.
Failure layers are separated.
CorrectLink retrieved evidence to final answers and record latency, cost, and refusals
End-to-end tradeoffs are comparable.
End-to-end tradeoffs are comparable.
IncorrectUse final answer accuracy alone to identify retrieval causes
Layers cannot be separated.
Layers cannot be separated.
IncorrectSkip evidence checks when retrieval looks plausible
Generation errors remain.
Generation errors remain.
Try it yourself
An example you can run in a temporary verification environment.
AWS公式AIF-C01 Domain 3.4のRAG評価、検索指標、生成品質を確認する。Expected result
検索と生成のどの段階が失敗したかを分離して改善できる。Key points
- RAG evaluation
- Retrieval
- Generation
Notes
- Environment: AWS公式AIF-C01試験ガイドとAWS公式ドキュメントの確認
- Command output formatting can vary slightly by distribution or tool version.
- Run the example in a temporary directory or process when possible.
Foundation review
Read the scope first
Check whether the command acts on the current shell, a new process, an existing process, or a file.
Verify the observable result
Use the supplied command and compare the output with the expected result.
Question IDAWS-AIF-443
EnvironmentAWS公式AIF-C01試験ガイドとAWS公式ドキュメントの確認
Last technical review2026-08-20
Report an error or rights issue