There is a much easier and safer way.. Program it to send and respond to the same signal. Send it back 100years to remain in place a few minutes and then return. Let us say you have it well calibrated to 10years but do not know exactly how well, which is why million year excursions would multiply your error and result in drift. So have the machine repeatedly send itself back 10 years 10 times. If the machine does not find itself after say 24 hours, return and assume the drift is under 10 years. Recalibrate to lengthen its concept of time and repeat. If the machine finds itself, note the delay before the first machine appears and calibrate accordingly. Repeat for 1000 years with 10 steps of one hundred. Do this then for 10000, 100000 and then one million.

There is a much easier and safer way.. Program it to send and respond to the same signal. Send it back 100years to remain in place a few minutes and then return. Let us say you have it well calibrated to 10years but do not know exactly how well, which is why million year excursions would multiply your error and result in drift. So have the machine repeatedly send itself back 10 years 10 times. If the machine does not find itself after say 24 hours, return and assume the drift is under 10 years. Recalibrate to lengthen its concept of time and repeat. If the machine finds itself, note the delay before the first machine appears and calibrate accordingly. Repeat for 1000 years with 10 steps of one hundred. Do this then for 10000, 100000 and then one million. BONUS:The method works for future calibration too.