I don't think anyone claims it's possible to read said data using standard drive firmware, so comparing it to uninitialised memory isn't the whole story. I always figured the theory was that if the cell was magnetised twice, you could tell from the magnitude of the resulting field not only its current magnetisation but also its history. I can see how that might be possible based on magnetic properties of real materials. ( http://en.wikipedia.org/wiki/Hysteresis_loop#Magnetic_hyster... ) However, you'd have to take an analog reading, not a binary one. The idea here being that N, then S magnetisation ought to yield a weaker S magnetisation than S, S.
Strong enough magnetisation will erase that history though, and presumably make the current data more long-lived and random bit-flips rarer. I'm just wondering if the tech used back in the days wasn't sophisticated to magnetise cells strongly enough. (without affecting neighbouring cells)
Yes, I probably am thinking about this too much. I guess that serves me right for doing a physics degree at university. :)
Imagination stems from trying to read from uninitialized memory, yielding an undefined value. :)