Peak shift
When flux transitions crowd together, adjacent readback pulses overlap and shift the peak in time — a bit read a little early or late.
At high density in MFM and RLL encoding, flux transitions sit close enough that their readback pulses (essentially Lorentzian curves) add together. The sum drags a pulse's peak away from its true position, and because the decoder hunts for that peak it clocks the bit at the wrong instant. The effect is also called pulse crowding or bit shift, and it bites hardest on the inner cylinders, where each revolution covers the least surface.
The fix is write precompensation: the controller writes the vulnerable transitions slightly early or late so they land correctly once shifted. RLL codes also cap how closely transitions may follow one another, which blunts the effect.