Comment by chii

> you've got someone who can judge if it's the system being stupid or if it's the test.

but why couldn't this be done even with the llm generated test cases?