Eckard,
You have hit one of the big nails, squarely on its head; "the necessity to integrate over future time too when analyzing past data."
I have pointed-out the problem with this previously, in the mathematical techniques at the foundation of QM, namely the use of Fourier Transforms, that integrate over all of time. How can one integrate over all of time, if one does not know the future?
Well, one can indeed know the future, for systems devoid of information, the very systems at the heart of classical physics. It is easy to predict the future of a conserved (constant) quantity, and it is easy to predict the future of a perfectly periodic function (idealized orbits). So, in those cases, one can indeed integrate over the future values, by integrating over the predictions.
Unfortunately, this does not work for unpredictable, high-information-content phenomenon, such as human observers. Unwittingly assuming that it does is THE problem. But this fact is not apparent in the Fourier transform formulation, and, consequently, has yet to be appreciated as a central problem in the mathematical formulation of QM, when it attempts to make claims about how observers behave and impact observable results..
Rob McEachern