What is artificial intelligence trajectory evaluation?
Definition
Artificial intelligence trajectory evaluation examines how an agent reached a result. It can measure model calls, tool use, intermediate decisions, corrections and whether the agent followed an efficient and safe path.
Trajectory evidence helps explain why a final answer succeeded or failed. Teams can compare changes to the model, harness and evaluator and monitor whether improvement in one metric introduces regressions elsewhere.
Acronyms and aliases
AI trajectory evaluation acronymagent trajectory evaluation variant
Related terms
Frequently asked questions
Why evaluate an agent's full trajectory?
A correct final answer can hide unsafe or inefficient steps, while a failed result can reveal a specific action that needs improvement.
What can a trajectory evaluator score?
It can score tool selection, action order, policy compliance, recovery, efficiency, intermediate reasoning artifacts and the final outcome.
