Does the second law of thermodynamics hold under non-equilibrium conditions? The question is just that: whether the second law of thermodynamics is always valid under non-equilinrium conditions?
The origin of life involves several chemical reactions that are thermodynamically not feasible $(+ \Delta G)$. To explain these reactions, non-equilibrium thermodynamics is invoked.  Hence, my question.
 A: If you define the system consistently, the 2nd law always holds. If you define the system inconsistently, you can violate any law of physics you like.
Case 1: The system is always the refrigerator in isolation, before, during, and after the thought experiment. All the normal laws of physics apply, but nothing happens.
Case 2: The system is the refrigerator and the room and the power plant and the electrical grid and the river that cools the power plant and the atmosphere of the planet, before, during, and after the thought experiment. All the normal laws of physics apply, interesting things happen, but the model is too complex to be useful.
Case 3: The system is the refrigerator in isolation before we run the thought experiment, work and heat magically enter and leave the system during the thought experiment, and then at the end of the thought experiment we just look at the refrigerator in isolation again. It's useful to set up the problem this way, packing all the unknown behavior of a complicated system into something simple and controllable like "work across the boundary". But we can't draw any conclusions about the laws of physics based on the beginning and ending states of "the refrigerator in isolation", because all the interesting physics stuff happened in a totally different system that includes wherever the work and heat came from and went to.
If we were being ontologically correct, we shouldn't ever use case 3. Instead...
Case 4: The system is always the refrigerator, a power supply thermally isolated from the refrigerator, and a heat sink in thermal contact with the refrigerator. All the normal laws of physics apply, interesting things happen, and we can model them.

To reiterate: if you take option 3 and try to infer laws of physics based on the start and end states of an inconsistently isolated system, you can violate whatever you want.
The system is a box with a ball rolling around in it. I allow magic transport of stuff across the boundary (I take the ball out of the box) and then I look at the box again. The ball is gone! The Second Law, gone! The First Law, gone! Newton's laws of motion, gone! Quantum mechanics, gone!
