I know you're being facetious, but I think this is correct. The model might ask for clarification when given clearly borderline questions that tread the line between what is true, what is false, and even what is misleading. But there's the rub of someone being disingenious and saying "no explanation! Just answer!" It was a trap to begin with.

I don't think there is anything wrong with the results of this test.

It would be more interesting if we compared them to human results.

If you have trouble distinguishing between human and LLM results, that's interesting.

Also, sentient is irrelevant to this test.