If the wrong level datum was input into all 3 systems then I would think the error would be the same as 'g' is incorrectly coupled into the accelerometer channel. Schuler tuning should give this an 84 minute or so period (from memory) but perhaps the observation wasn't long enough to observe. In any case, I would have thought that the system will be correcting against real time GPS with the Kalman filter estimating both position & dynamics, and also estimating the error characteristics of the accelerometers and gyros. So all should have been good but with some unusally high errors in whatever channel was compromised (eg accelerometer bias).