Testing of LLM applications should be at the expected behavioural level, not at the implementation level. This means that we should focus on the expected outputs of the application, rather than the ...