Code Execution Metric

What is Code Execution Metric?

A code execution metric scores generated code by running it against tests or expected outputs—pass@k style—rather than only comparing text to a reference. It measures whether the program works, not whether it looks similar.

Related Giskard articles

Test executable agent code with Giskard — pair execution metrics with tool-misuse probes when agents can run code. See function-calling security.

Authority: OpenAI HumanEval / pass@k (arXiv).

Get AI security insights in your inbox