Agent trajectories: success, efficiency, recovery, and policy
An agentic trajectory is evaluated by outcome and process: tools and observed states explain how the result was reached, while recovery, cost, and policy separate clean success from corrupt success.
Links containing ?t= open the video at a specific second.
Video summary
The ideas to retain
A trajectory is a sequence of transitions, not a list of tool names
For reasoning about one execution, represent each step as:
Start with the outcome: was the task actually solved?
When the result can be checked directly, the first grader should inspect the real state.
Then evaluate the decisions that matter along the way
Not every trajectory difference is important.
Key moments
Jump directly to a section
Next step
