QABattle

Agentic AI Testing / Quiz

Evaluating Tool-Call Trajectories

The final answer was right, but the agent took six wrong turns to get there. Outcome-only evals miss it. Trajectory evals don't.

Difficulty
Medium
Format
Quiz
Points
150
Estimate
10 min

// MISSION BRIEF

Your Mission

An agent's output is one thing; the PATH it took, the sequence of tool calls, observations, and decisions, is another. Trajectory evaluation grades the path. This quiz covers what to measure and why outcome-only scoring hides real defects.

Scoring: equal weight per question.

// FIRST CONTACT

Battle teaser

In agent evaluation, a 'trajectory' is:

  1. AThe ordered sequence of steps the agent took, tool calls, arguments, observations, and intermediate decisions
  2. BThe system prompt
  3. CThe model's temperature setting
  4. DThe final answer string only
Answers, scoring, hints, and the full battle stay sealed.

// SKILL TAGS

agentictrajectory-evaltool-callingevaluation