I think your example is a case where it would work well. Imagine Party A and Party B both produce models to estimate the effects of tax changes on job growth.

I think your example is a case where it would work well. Imagine Party A and Party B both produce models to estimate the effects of tax changes on job growth. Based on these models each Party produces a policy proposal. Party A is in power so their proposal is chosen. This is now a natural experiment to test both models! Data collected after the policy is implemented can now be used to judge the validity of both models. If a model is shown to be wrong, the Party either needs to explain the inconsistency or change their model to suit. Why doesn't this occur in the real world? Politicians know that many of their policies are not predicated on fact but on pragmatic political considerations. Revealing their ignorance by publishing models is just adding to the attack surface that their opponents can use to criticize them. Thus as long as there are no competing parties producing explainable (in the modelling sense) policy, there is no advantage to explaining policy in this way, it is all downside. It would take a very courageous group to change the paradigm, and they would probably be ridiculed because they would need to reveal their ignorance when publishing their models.