When solving a policy / strategic problem - I have typically thought about exploring the different policy options as a scenario planning test where policy options are different inputs. Is this how others think about this? Or would you solve for the right inputs as its own optimization? i.e. if I’m trying to solve for the best place to put a recharge station for an EV, should the possible recharge stations be varied in the input and run through a scenario test? Or should I build an optimization model which takes in the demand for charging stations and outputs the optimal locations on a map? And if the answer is the second option - how would you do this? TL;DR - should I scenario test by varying inputs to a model or by modifying the model to do it for me?
What do the different policies correspond to? Mentioning that this is strategic makes me believe you have a general business problem to solve, and there might be multiple pieces to that solution.
So in this case is a policy one of your inputs? It sounds like you’re looking to solve two problems: (1) you have a flow problem (how many EV stations do I need and where?), and (2) you have a routing problem – where you’re looking for some sequence of EV station stops along your route.
If I understand correctly, I’d probably split it into the two solves. One is a flow problem where you’re optimizing for ideal positioning and number of nodes. In my opinion, this can be an entire analysis on its own. I’d use the result of that analysis/solution as inputs for the routing model.
Curious what others think about this one. 
I have typically thought about exploring the different policy options as a scenario planning test where policy options are different inputs.
With regards to testing methodology, you can go both directions and each helps you deal with a different type of uncertainty. For instance, you can run such a test with fixed inputs searching through the parameter space of a given model (or the formulational space) to understand model uncertainty, and you can vary the inputs with a fixed model to deal with state uncertainty.
I’m trying to solve for the best place to put a recharge station for an EV, should the possible recharge stations be varied in the input and run through a scenario test?
I second @cnpryer here. This sounds like it warrants a decomposition approach. If we combine these decisions into a single model, the state space will likely be so large that no solver can search it effectively. If we fix certain first-stage decisions, in this case where to put EV charging stations, then we can solve the downstream problem of how to route vehicles under those conditions.
The trick is figuring out how to estimate the downstream impacts of those first-stage decisions. One option is to create a set of scenarios, either synthetic or pulled from history, and use those in the outer model. Another is to use something like Benders.