(Disregarding a few complexities about the quality and stability of the models)

The problem of linking different "facts" and actions is ancient and has a plain solution: the politician applies the weighs of different interests according to the relative power behind those interests (including tactical or strategical considerations about the consequences). I am not sure that to call those interests «moral values» is actually healthy.

One issue we cannot disregard is that not only facts can be tentative: the main issue we have is with tentative solutions - it is more with guessing which solution will be effective (its estimated consequences get weighed according to the interests in order to rank the proposal) than to decide which interpretation of the world is more adequate.

Edit: anyway, Bratton seems to be fully aware and explicit in considering that "fact are less of an issue than solutions" and that "said solutions are to be calibrated subjectively, politically - since an "objective", all-encompassing algorithm is out of reach".