I interned at the Charles Stark Draper Laboratory 20+ years ago. One of the guys was telling me at lunch about his work that day. He had to run some simulations to tell the Navy if their Trident guidance system needed to be sent back for re-manufacturing after a forklift operator dropped it. My understanding is that the handling crate had some kind of integral rough handling detector. (I presume calibrated weights on thin wires, in 3 axes, so you can bracket the peak accelerations by looking at which wires broke and which didn't.)
I was told that once above most of the atmosphere, the MIRV bus guidance unit pops the lens off and takes a snapshot of the stars. The MIRV bus is rotating at that point, so the telescope scans a good section of the sky, but it takes a snapshot at a very specific time and is only interested in a rather small patch of sky. It compares what it sees with a stored copy of what it should see, and uses that to re-calibrate the inertial measurement unit.
The way it was described to me was that one rotation checked the star field, the IMU was re-calibrated, the very next rotation was used to double-check the IMU calibration, and right after that, the MIRV bus started popping off the individual warheads. It seems strange to me that they would use such a short observation window to re-calibrate the IMU, rather than having a long period of continuous re-calibration. It's also highly possible that the guy explaining it to me was giving intentional misinformation to avoid accidentally disclosing TS/SCI to a non-cleared intern.
The nitty-gritty of guidance is fascinating. It's unfortunate that the main use cases for ultra-accurate guidance are in weapons.
There was a pendulous integrating gyroscopic accelerometer from the Apollo IMU sitting in a glass case in the lab... a shame given how few people were allowed into the lab.