jargon

Applied AI·Agents

you replay the whole recorded run, thought by tool call by result, to find the step where it went wrong.

Trajectory

Draft summary, pending review

The complete recorded sequence of one agent run: every thought, tool call, result and message. The unit of agent debugging and evaluation; you replay and score trajectories the way you replay failing test cases.