Multi-Turn & Agentic Evals
Evaluate multi-turn conversations and agentic systems, including HITL scripts and tool-use trajectories.
Agentic evals cover systems that reason across multiple turns, invoke tools, or require human-in-the-loop scripts to drive the conversation.
What you will learn
How to define a multi-turn eval with a HITL script.
How trajectory scorers evaluate tool-use sequences.
Patterns for evaluating Harness Agent V3 and similar systems.
Last updated
Was this helpful?