Thanks so much, that’s very clear and helpful!
I understand that it’s not possible to identify effects entirely independently, but I’ll look into getting the average marginal effect or something similar. There’s this thread on here, and there are quite some resources for R (like this one, and it’s even part of brms, but sadly nothing similar exists for python/pymc as far as I can see). I suppose that’s the best estimate of ‘independent’ effects that I can get, which I can then interpret cautiously as you suggest.