After PR3 alignment by Ikeda-san and Yuzurihara-kun, SDF differences appeared on k1calcs,
which came from a numerical rounding error between the actual EPICS records and set point
values in the snapshot file.
I called Miyoki-san and reverted differences shown in Fig.1.

-----
We need to do following procedure to accept new values which are computed by some kind of scripts.
1. accept differences in observation.snap
2. load an another table such as down.snap
3. reload observation.snap
4. revert differences in observation.snap

Because I forgot to do 2-4 in the maintenance day and 2 and 3 were done during the alignment work,
(When DOWN is requested, down.snap is loaded. And also in LOCK_PREP, observation.snap is (re-)loaded.)
differences coming from the numerical rounding errors appeared.
In the case of normal lockloss, DOWN is not requested. So this issue occurs at the first time of DOWN request.

Though this issue can be mitigated by the proper digit rounding,
the digit rounding doesn't cover all cases.
It's depending on the binary notation of values.
Anyway, we must not forget the procedure of 1-4. It's just my mistake in this time, I'm sorry.