Essentially the IRS is carrying two positions - one which is the product of it's computation from the accelerometers (based on its initialisation position) and one derived from the updates received from wherever (FMS etc.). These two positions are inevitably going to be different over time due to IRS drift and the amount of that drift can be resolved with the aircraft parked at an accurately known ground position at the end of flight. The acceptable amount of drift differs between installations but, as an example, the L1011 computation was based on a maximum acceptable error of 3 + 3T in nautical miles, where T was hours from initialisation.