It's not clear to me that what this kind of experiment does is test decisions. It certainly tests something, which may be what I would call the neurological precursor of action. I don't think decisions are always reduced to that. I may take the decision to attend a meeting two months in advance of it and then change my mind and take a different decision one week before the meeting and change again at the last moment. Does this device tests that?Why would the device wait? If the result is determined by prior factors such as brain state, environmental conditions, and so on, then it will know what you're going to do well before you do it, in plenty of time for you to press either button. Indeed, if your decision to press is determined, that decision must come before your action to press, with enough time for you to initiate either action.
EB
If you're referring to Libet-style experiments, then I agree with your criticisms and share your reservations.
However, this isn't one. It's simply a thought experiment, in which you have a box (Godelbox) that scans the relevant environment, including your brain states, for the precursors of decisions and actions that people say determine what you will do in terms of button pressing. And then works out what button you will press. And the lights up so that you know which button it will be.
The point is to illustrate a logical problem with determinism.