Forget the past and design a test that involves deja vu?It depends on what you are predicting and when. If you are telling someone about it after the event and you are the only evidence then that doesn't count obviously. Randi had a method that would have worked but you felt that it was asking "too many questions".
If you are predicting whether a coin will be heads of tails then if you can call it more than 50% of the time correctly given a large enough sample that works.
Predicting David's dream in your dream and then having only you or David give the feedback isn't going to pass any scientific test.
Like if I have a dream of a scene overnight and it is written down first thing in the morning, and when the deja vu experience happens I record it somehow, and you then compare the dream vision (notes) to the situation (photograph?).
Can you see how a test can be devised to prove or disprove that?