The RG used in statistical physics also suffers from being ad-hoc. There is a constant stream of physics papers that propose a new "architecture" for doing RG a

The RG used in statistical physics also suffers from being ad-hoc. There is a constant stream of physics papers that propose a new "architecture" for doing RG and comparing the numerics to other architectures, with very little (it seems) understanding of why these things do or do not work as well. For example: https://arxiv.org/abs/1412.0732 https://arxiv.org/abs/1412.0732