So I had to reconstruct a signal that fitted the data for the equations to work at all. But the reconstruction was no longer measured data. It was made up, and there was no single solution for fitting the data. That meant I could claim to "respect the data" while producing any result I wanted.