LST calculated from the observer's location and the time is ~20 s later than the LST logged during the observation?

[TakKanekoGit]

Based on observation data provided for the "roll issue" (3f7cc690 at 2024-05-02T03:39:20.312763+00:00), the LST calculated from the observer's location and the time in the log seems to be 20 seconds later than the LST logged during the observation. If this difference is real, it would corresond to a 5-arcmin position discrepancy.

Details and examples in the unit test: unit_test_calc_utils.test_sf_get_lst_hrs