High error threshhold in getAdaptiveGain

[FinnMeyer]

This error function in getAdaptiveGain
double a_mag = sqrt(ax * ax + ay * ay + az * az);
double error = fabs(a_mag - kGravity) / kGravity;
allows very high ax or ay accelerations before shutting down. E.g. you can have an ax of 6.5 m/s^2 before the error reaches 0.2
and the adaptive gain is completly cut.
In the checkState the max error is defined as:
fabs(acc_magnitude - kGravity) > kAccelerationThreshold
without the / kGravity;
shouldn't this also be the case here?