'it's not just a feature of my or anyone's "model" of the world.'
This may just be semantics, but I dont see how you could argue that it isnt a feature of a model of the world. A neural representation of the world shared amongst a group of humans. Tested against reality in the best ways we can imagine. But it is still just a model, it is not actual reality. Are you really so sure about those priors? You are also have a very influential prior there which supposes that that if the device functions, it does so because it violates conservation of momentum. Surely you cant know this? Are there really no other explanations which fit more neatly with our current models?