This kind of thing is inherently un-provable. You can have a method that works stupidly well for decades, and then it stops working. Building a model that provenly (in the mathematical sense) works with X% efficiency would be akin to building a free energy device.