The very brief version is: a fully pressurized tank squishes the Belleville washers, pushing the orifice toward the seat on the end of the piston. That slightly squishes the piston spring, allowing it to push back a little harder. That decreases the IP lower than it would be if the orifice hadn’t moved. That lowers the IP pressure you would have with a full tank relative to a system without the moving orifice.
As the tanks pressure decreases, that effect lessens, meaning that the IP would be slightly higher. But that is offset by the fact that the pressure of the tank is lower, which means it’s pushing against the seat a little less, which would normally cause the IP pressure to drop a little. That keeps the IP closer to the same pressure it was when it was a full tank. Theoretically, that continues the entire way down as the tank pressure drops.
It’s not perfect, but it greatly decreases the amount of IP change between a full tank and an empty tank.
So instead of keeping the tank force and the piston force perpendicular to each other, with full tank pressure on all sides of the piston and very carefully machining everything to minimize imbalances like you have in a truly balanced first stage, you have a traditional unbalanced first stage that mechanically offsets the pressure shift inherent in an unbalanced first stage. Simple, clever, easy to produce… Not as good. But it has a certain appeal of its own.