5.1.4 · Lesson 4 of 14
Validate Retrieval and Agent Task Success
Measure retrieval quality and autonomous agent effectiveness across multi-step tasks and tool use.
WHAT
Validate retrieval success and multi-step agent workflow effectiveness.
HOW
- Task Completion Rate: % of multi-step goals solved correctly.
- Tool Usage Efficiency: valid tool calls vs hallucinated/failed calls.
- Reasoning Path Quality: review traces for logical loops.
- For RAG, score context relevance alongside answer metrics.
Key points
- Task Completion Rate
- Tool Usage Efficiency
- Reasoning Path Quality
- Context precision/recall
- Trace log review
Exam focus
Agent evaluation tracks task completion, valid vs hallucinated tool calls, and reasoning-path quality — not only final answer text.